Aerodynamic configuration optimization method, device and equipment for complex object and medium

Abstract

The invention provides an aerodynamic configuration optimization method and device for a complex object, equipment and a medium, and the method comprises the steps: carrying out the displacement parameter transformation of a plurality of model points of an initial three-dimensional geometric model of the object, and carrying out the deformation processing of the initial three-dimensional geometric model through combining with a direct free deformation technology, so as to generate a new three-dimensional geometric model; performing aerodynamic performance prediction processing on the point cloud data of the plurality of new three-dimensional geometric models based on an aerodynamic performance prediction model, and outputting aerodynamic performance parameters of each new three-dimensional geometric model; if a plurality of aerodynamic performance parameters of the new three-dimensional geometric model exist, searching and selecting a plurality of non-dominated optimal solutions from the plurality of new three-dimensional geometric models based on non-dominated sorting and a reference point guiding mechanism, and dynamically adjusting displacement parameters of model points based on clustering analysis results of the plurality of non-dominated optimal solutions; and when a preset convergence condition is met, outputting an optimal aerodynamic configuration design scheme. And the optimal aerodynamic configuration design scheme of the object can be quickly and accurately generated.

Description

Technical Field

[0001] The present application relates to the fields of aerodynamics and object shape optimization technology, and in particular to a method, device, equipment and medium for optimizing the aerodynamic shape of a complex object. Background Art

[0002] With the increasing demand for industrial design and engineering optimization, the shape optimization of complex objects (such as cars, airplanes, ships, etc.) has become a key link in improving performance and energy efficiency. Traditional shape optimization methods mostly rely on manual iteration, parametric modeling, and numerical simulation technologies such as finite element and finite volume. Although they have a certain degree of reliability and maturity, the process is highly dependent on manual experience, with low overall efficiency and long cycle time. It is difficult to support high-frequency, high-dimensional parameter deformation analysis and rapid iterative optimization. In addition, when faced with complex nonlinear design spaces or multi-objective trade-offs, traditional methods have obvious limitations in searching for global optimal solutions, making it difficult to meet the needs of modern engineering for efficient and intelligent optimization. Therefore, how to improve the accuracy of aerodynamic shape optimization of complex objects has become a technical issue that cannot be underestimated. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method, device, equipment and medium for optimizing the aerodynamic shape of a complex object. Through direct free deformation technology, aerodynamic performance prediction model and intelligent optimization processing for multiple aerodynamic performance parameters, the optimal aerodynamic shape design scheme of the object can be generated quickly and accurately.

[0004] The present invention provides a method for optimizing the aerodynamic shape of a complex object. The method includes: Performing displacement parameter transformation on multiple model points of the initial three-dimensional geometric model of the object, and deforming the initial three-dimensional geometric model using direct free-form deformation technology to generate multiple new three-dimensional geometric models; performing aerodynamic performance prediction processing on the point cloud data of the plurality of new three-dimensional geometric models based on the aerodynamic performance prediction model, and outputting aerodynamic performance parameters of each of the new three-dimensional geometric models; If there are multiple aerodynamic performance parameters for the new three-dimensional geometric model, multiple non-dominated optimal solutions are searched and selected from the multiple new three-dimensional geometric models based on the non-dominated sorting and reference point guidance mechanism. The displacement parameters of the model points are dynamically adjusted based on the clustering analysis results of the multiple non-dominated optimal solutions. When the preset convergence conditions are met, the optimal aerodynamic shape design scheme of the object is output.

[0005] In a possible implementation manner, after outputting the aerodynamic performance parameters of each of the new three-dimensional geometric models, the aerodynamic shape optimization method further includes: If the aerodynamic performance parameter of the new three-dimensional geometric model is one, the displacement parameters of the model points corresponding to the multiple new three-dimensional geometric models are iteratively optimized based on the genetic algorithm, and the optimal aerodynamic shape design scheme of the object is output when the preset convergence condition is met.

[0006] In one possible implementation, for any of the new three-dimensional geometric models, performing aerodynamic performance prediction processing on the point cloud data of multiple new three-dimensional geometric models based on the aerodynamic performance prediction model and outputting aerodynamic performance parameters of each new three-dimensional geometric model includes: Performing feature extraction and encoding processing on the new three-dimensional geometric model in the aerodynamic performance prediction model to determine voxel features corresponding to each point cloud data; Processing the plurality of voxel features based on a convolutional layer, a linear layer, a normalization layer, and an activation function, and outputting a plurality of embedded features; Processing the plurality of embedded features in sequence based on a local attention layer, a feedforward network layer, a residual connection layer, and a normalization layer in a multi-layer encoder to output a plurality of multi-scale features; Global pooling is performed on the plurality of multi-scale features to output global features, and mapping is performed on the global features based on a regressor to output the aerodynamic performance parameters.

[0007] In a possible implementation, searching and selecting multiple non-dominated optimal solutions in multiple new three-dimensional geometric models based on the non-dominated sorting and reference point guidance mechanism includes: A population is formed based on the multiple new three-dimensional geometric models and the corresponding aerodynamic performance parameters, and all individuals in the population are sorted according to non-domination relationships to generate a non-dominated hierarchy; wherein the first level of the non-dominated hierarchy is a completely non-dominated solution set, and the second level is a non-dominated solution set remaining after removing the first level; A set of reference points are evenly arranged in the target optimization space, and when selecting the next generation of individuals from the population, individuals with high non-dominated levels close to the reference points are preferentially selected, and the individuals with high non-dominated levels are used as the non-dominated optimal solutions; wherein the reference points are optimization targets corresponding to multiple aerodynamic performance parameters.

[0008] In one possible implementation, dynamically adjusting the displacement parameters of the model points based on the cluster analysis results of the plurality of non-dominated optimal solutions and outputting the optimal aerodynamic shape design solution for the object when a preset convergence condition is met includes: performing cluster analysis on the plurality of non-dominated optimal solutions to identify solution clusters with different performance tendencies; Directedly shrink the spatial range of the displacement parameters of the model points of each solution cluster. In the next generation of evolution, according to the distribution density of each solution cluster in the target space, continue to use the reference point mechanism and adaptive population allocation strategy to dynamically adjust the distribution ratio of the population among the clusters. When the aerodynamic performance parameters of the solution cluster meet the preset convergence conditions, the solution cluster is used as the optimal aerodynamic shape design solution for the object.

[0009] In one possible implementation, the iterative optimization of displacement parameters of the model points corresponding to the plurality of new three-dimensional geometric models based on a genetic algorithm, and outputting an optimal aerodynamic shape design scheme for the object when a preset convergence condition is met, includes: Using the displacement parameters of the plurality of model points as the initial population in the genetic algorithm; Taking the aerodynamic performance parameters as fitness functions, using a genetic algorithm to perform multi-generation iterative optimization on the population, generating a new generation of individuals in each generation, and retaining individuals with high fitness as the optimal solution; Determine a search interval for the displacement parameters based on the distribution of the displacement parameters of the model points of the optimal solution in the current generation, and expand a buffer boundary of a preset proportion based on the search interval; When the individual aerodynamic performance parameters corresponding to the displacement parameters of the model points within the search interval meet the preset convergence conditions, the individual and its corresponding three-dimensional geometric model and aerodynamic performance parameters are used as the optimal aerodynamic shape design scheme.

[0010] In a possible implementation manner, the aerodynamic performance prediction model is determined by the following steps: Inputting sample point cloud data of multiple sample three-dimensional geometric models and corresponding sample aerodynamic performance parameters into a deep learning network model to learn the mapping relationship between the sample point cloud data and the sample aerodynamic performance parameters, and predicting the predicted aerodynamic performance parameters of the sample point cloud data; The deep learning network model is iteratively trained based on the loss value between the predicted aerodynamic performance parameters and the sample aerodynamic performance parameters to determine the aerodynamic performance prediction model.

[0011] The present application also provides an aerodynamic shape optimization device for a complex object, the aerodynamic shape optimization device comprising: A free deformation processing module is used to transform the displacement parameters of multiple model points of the initial three-dimensional geometric model of the object, and then deform the initial three-dimensional geometric model using direct free deformation technology to generate multiple new three-dimensional geometric models; an aerodynamic performance prediction module, configured to perform aerodynamic performance prediction processing on the point cloud data of the plurality of new three-dimensional geometric models based on an aerodynamic performance prediction model, and output aerodynamic performance parameters of each of the new three-dimensional geometric models; An intelligent optimization module is used to search and select multiple non-dominated optimal solutions from the multiple new three-dimensional geometric models based on a non-dominated sorting and reference point guidance mechanism if there are multiple aerodynamic performance parameters of the new three-dimensional geometric model, dynamically adjust the displacement parameters of the model points based on the clustering analysis results of the multiple non-dominated optimal solutions, and output the optimal aerodynamic shape design scheme of the object when the preset convergence conditions are met.

[0012] An embodiment of the present application also provides an electronic device, comprising: a processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus, and when the machine-readable instructions are executed by the processor, the steps of the aerodynamic shape optimization method of a complex object as described above are performed.

[0013] An embodiment of the present application further provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the aerodynamic shape optimization method of a complex object as described above are executed.

[0014] Embodiments of the present application provide a method, apparatus, device, and medium for optimizing the aerodynamic shape of a complex object. The method comprises: performing displacement parameter transformation on multiple model points of an initial three-dimensional geometric model of the object, then deforming the initial three-dimensional geometric model using a direct free-form deformation technique to generate multiple new three-dimensional geometric models; performing aerodynamic performance prediction on the point cloud data of the multiple new three-dimensional geometric models based on an aerodynamic performance prediction model, and outputting aerodynamic performance parameters for each of the new three-dimensional geometric models; if the new three-dimensional geometric models have multiple aerodynamic performance parameters, searching and selecting multiple non-dominated optimal solutions from the multiple new three-dimensional geometric models based on a non-dominated sorting and reference point guidance mechanism; dynamically adjusting the displacement parameters of the model points based on cluster analysis results of the multiple non-dominated optimal solutions; and outputting an optimal aerodynamic shape design solution for the object when a preset convergence condition is met. By using the direct free-form deformation technique, the aerodynamic performance prediction model, and the intelligent optimization process for multiple aerodynamic performance parameters, an optimal aerodynamic shape design solution for the object can be quickly and accurately generated.

[0015] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other relevant drawings can be obtained based on these drawings without creative work.

[0017] Figure 1 A flowchart of a method for optimizing the aerodynamic shape of a complex object provided in an embodiment of the present application; Figure 2 A schematic diagram of a method for optimizing the aerodynamic shape of a complex object provided in an embodiment of the present application; Figure 3 This is one of the structural schematic diagrams of a device for optimizing the aerodynamic shape of a complex object provided in an embodiment of the present application; Figure 4 This is a second structural diagram of a device for optimizing the aerodynamic shape of a complex object provided in an embodiment of the present application; Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for which protection is claimed, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, each other embodiment obtained by those skilled in the art without making creative work falls within the scope of protection of the present application.

[0019] First, the application scenarios to which this application is applicable are introduced. This application can be applied in the fields of aerodynamics and object shape optimization technology.

[0020] Research has found that with the increasing demand for industrial design and engineering optimization, the shape optimization of complex objects (such as cars, airplanes, ships, etc.) has become a key link in improving performance and energy efficiency. Traditional shape optimization methods mostly rely on manual iteration, parametric modeling, and numerical simulation technologies such as finite element and finite volume. Although they have a certain degree of reliability and maturity, the process is highly dependent on manual experience, with low overall efficiency and long cycles. It is difficult to support high-frequency, high-dimensional parameter deformation analysis and rapid iterative optimization. In addition, when faced with complex nonlinear design spaces or multi-objective trade-offs, traditional methods have obvious limitations in searching for global optimal solutions, making it difficult to meet the needs of modern engineering for efficient and intelligent optimization. Therefore, how to improve the accuracy of aerodynamic shape optimization of complex objects has become a technical issue that cannot be underestimated.

[0021] Based on this, an embodiment of the present application provides a method for optimizing the aerodynamic shape of a complex object. Through direct free deformation technology, an aerodynamic performance prediction model, and an intelligent optimization processing process for multiple aerodynamic performance parameters, the optimal aerodynamic shape design scheme of the object can be generated quickly and accurately.

[0022] See also Figure 1 , Figure 1 This is a flow chart of a method for optimizing the aerodynamic shape of a complex object provided in an embodiment of the present application. Figure 1 As shown in , the aerodynamic shape optimization method provided by the embodiment of the present application includes: S101: After performing displacement parameter transformation on multiple model points of an initial three-dimensional geometric model of an object, the initial three-dimensional geometric model is deformed using a direct free-form deformation technique to generate multiple new three-dimensional geometric models.

[0023] It should be noted that multiple control frames are set on the initial three-dimensional geometric model, the model points within the control frames are transformed by displacement parameters, and then the direct free deformation technology is used to deform the initial three-dimensional geometric model after the displacement parameter transformation to generate a new three-dimensional geometric model.

[0024] Specifically, an initial 3D geometric model (such as a DrivAer car model, a SUBOFF submarine model, or a CHN-T1 aircraft, among other standard aerodynamic shapes) is selected. A bounding box (BoundingBox) that completely encloses the model is automatically calculated, and its dimensions (length × width × height) are determined. A grid of control points (e.g., 11 × 9 × 9) is set in the x, y, and z directions of the bounding box to form a direct free-form deformation (DFFD) framework. Deformation regions are selected: model points in key areas of the model surface (e.g., the front of a car, the tail wing, the bow of a submarine, etc.) are divided and their allowable displacement directions are set (e.g., movement in the x and z directions only). Latin hypercube sampling (LHS) is used to generate 200 or more deformation parameter combinations to ensure uniform coverage of the design space. The DFFD-driven model is then used to deform the model points corresponding to each parameter set to generate a new 3D geometric model.

[0025] S102: performing aerodynamic performance prediction processing on the point cloud data of the multiple new three-dimensional geometric models based on the aerodynamic performance prediction model, and outputting aerodynamic performance parameters of each of the new three-dimensional geometric models.

[0026] In this step, an aerodynamic performance prediction model is used to perform aerodynamic performance prediction processing on the point cloud data of multiple new three-dimensional geometric models, and the aerodynamic performance parameters of each new three-dimensional geometric model are output.

[0027] It's important to note that the aerodynamic performance prediction model dynamically captures the distribution characteristics of point clouds in key areas (such as curvature changes, pressure gradients, and vortex core locations) through local geometric feature aggregation and multi-scale transformation. This allows for precise characterization of the impact of complex flow phenomena (such as separated flows and wake vortex structures) on aerodynamic performance. Compared to traditional gridded CFD data, the aerodynamic performance prediction model directly learns features from unstructured point clouds, making it more suitable for automated modeling of complex geometries.

[0028] Here, aerodynamic performance parameters include drag coefficient, lift coefficient, and pressure distribution, etc. The aerodynamic performance parameters corresponding to different complex objects are different, and are not specifically limited in this section.

[0029] In one possible implementation, for any of the new three-dimensional geometric models, performing aerodynamic performance prediction processing on the point cloud data of multiple new three-dimensional geometric models based on the aerodynamic performance prediction model and outputting aerodynamic performance parameters of each new three-dimensional geometric model includes: (1): Feature extraction and encoding processing are performed on the new three-dimensional geometric model in the aerodynamic performance prediction model to determine the voxel features corresponding to each point cloud data.

[0030] Here, we use PyTorch's Dataset and DataLoader, which support batch loading and multi-process acceleration. Each batch contains point cloud data for several new 3D geometric models. The continuous point cloud coordinates are discretized into integer grid coordinates (e.g., one grid per 1 cm). This facilitates subsequent spatial sorting and sparse convolution. The voxelized point cloud is spatially sequenced (e.g., using z-order or Hilbert curve encoding) so that spatially adjacent points are also as close as possible in the sequence, facilitating patch grouping and local attention. The point cloud data is then converted into a sparse tensor to obtain voxel features.

[0031] (2): Based on the convolution layer, linear layer, normalization layer and activation function, the multiple voxel features are processed to output multiple embedded features.

[0032] Here, a 3D convolution + linear layer is used to map the 3D coordinates of each voxel feature to a higher-dimensional feature space (such as 32 dimensions). Then, after normalization and activation function processing, the hidden layer features are obtained.

[0033] (3): Based on the local attention layer, feedforward network layer, residual connection layer and normalization layer in the multi-layer encoder, multiple embedded features are processed in sequence to output multiple multi-scale features.

[0034] Here, the local attention layer is used to perform self-attention within adjacent serialized patches, capturing spatial local relationships. The feedforward network layer (MLP) is used to enhance feature representation. The residual connection layer and normalization layer are used to improve training stability. With each layer, the feature dimension gradually increases (e.g., 32 → 64 → 128 → 256), while the number of points gradually decreases.

[0035] (4): Perform global pooling processing on the plurality of multi-scale features to output global features, perform mapping processing on the global features based on a regressor, and output the aerodynamic performance parameters.

[0036] Perform global pooling (such as average pooling) on ​​all current multi-scale features to obtain global features. The regressor maps the global features and outputs aerodynamic performance parameters.

[0037] In one possible implementation, the aerodynamic performance prediction model is determined by the following steps: The sample point cloud data of multiple sample three-dimensional geometric models and the corresponding sample aerodynamic performance parameters are input into a deep learning network model to learn the mapping relationship between the sample point cloud data and the sample aerodynamic performance parameters, and predict the predicted aerodynamic performance parameters of the sample point cloud data; the deep learning network model is iteratively trained based on the loss value between the predicted aerodynamic performance parameters and the sample aerodynamic performance parameters to determine the aerodynamic performance prediction model.

[0038] Here, if the loss value is greater than the preset loss value, the network parameters of the deep learning network model are updated, and the updated deep learning network model continues to be iteratively trained. If the loss value is less than or equal to the preset loss value, the current deep learning network model is used as the aerodynamic performance prediction model.

[0039] Among them, before inputting the sample data into the deep learning network model, the sample data needs to be preprocessed: data format processing, data standardization processing, downsampling and repeated sampling processing, etc.

[0040] It should be noted that the sample aerodynamic performance parameters are determined using the LBM algorithm. The LBM algorithm has strong localization and, when combined with GPU acceleration, can significantly improve computational efficiency, making it particularly suitable for rapid iterative optimization of multiple solutions. Furthermore, it automatically ensures mass conservation and offers flexible boundary handling (such as dynamic boundaries and curved surface boundaries). It can accurately capture complex flow characteristics such as separated flows and vortex shedding, providing high-fidelity predictions of aerodynamic forces (such as drag, lift, and lateral force) and flow field details for complex shapes such as aircraft, vehicles, or buildings. For example, for automobiles, the LBM method can rapidly calculate sample aerodynamic performance parameters corresponding to different sample 3D geometric models generated by DFFD technology.

[0041] S103: If there are multiple aerodynamic performance parameters for the new three-dimensional geometric model, multiple non-dominated optimal solutions are searched and selected from the multiple new three-dimensional geometric models based on the non-dominated sorting and reference point guidance mechanism, and the displacement parameters of the model points are dynamically adjusted based on the clustering analysis results of the multiple non-dominated optimal solutions. When the preset convergence conditions are met, the optimal aerodynamic shape design scheme of the object is output.

[0042] In this step, if there are multiple aerodynamic performance parameters for the new three-dimensional geometric model, multiple non-dominated optimal solutions are searched and selected from the multiple new three-dimensional geometric models according to the non-dominated sorting and reference point guidance mechanism. The displacement parameters of the model points are dynamically adjusted according to the clustering analysis results of the multiple non-dominated optimal solutions. When the preset convergence conditions are met, the optimal aerodynamic shape design scheme of the object is output.

[0043] It should be noted that the optimal aerodynamic shape design scheme includes a three-dimensional geometric model and corresponding aerodynamic performance parameters.

[0044] The convergence condition here is whether the aerodynamic performance parameters of the current 3D geometric model meet the preset standard aerodynamic performance parameters. For example, in a car drag optimization task, the task terminates when the drag coefficient is < 0.001 for five consecutive iterations or when the maximum number of iterations (500 by default) is reached.

[0045] In one possible implementation, searching and selecting multiple non-dominated optimal solutions in multiple new three-dimensional geometric models based on the non-dominated sorting and reference point guidance mechanism includes: A: A population is formed based on multiple new three-dimensional geometric models and corresponding aerodynamic performance parameters, and all individuals in the population are sorted according to non-dominated relationships to generate a non-dominated hierarchy; wherein the first level in the non-dominated hierarchy is the completely non-dominated solution set, and the second level is the non-dominated solution set remaining after removing the first level.

[0046] Here, a population is formed according to a plurality of new three-dimensional geometric models and corresponding aerodynamic performance parameters, and all individuals in the population are sorted according to the non-dominance relationship to generate a non-dominance level.

[0047] Among them, the first level in the non-dominated hierarchy is the completely non-dominated solution set, the second level is the non-dominated solution set remaining after removing the first level, and so on.

[0048] It should be noted that a solution is called “non-dominated” if it is not worse than another solution in all optimization objectives and is better than at least one optimization objective, where the optimization objective is the standard aerodynamic performance parameter.

[0049] A set of reference points are evenly arranged in the target optimization space, and when selecting the next generation of individuals from the population, individuals with high non-dominated levels close to the reference points are preferentially selected, and the individuals with high non-dominated levels are used as the non-dominated optimal solutions; wherein the reference points are optimization targets corresponding to multiple aerodynamic performance parameters.

[0050] Here, in the multi-objective optimization process, a set of "reference points" are pre-uniformly arranged in the target space (such as wind resistance, lift, wind noise, etc.). These reference points are used to guide the solution set to be evenly distributed in the high-dimensional target space. When selecting the next generation of individuals in the population, individuals with high non-dominated levels close to the reference points are given priority, and individuals with high non-dominated levels are regarded as non-dominated optimal solutions.

[0051] The process of determining individuals with high non-dominated levels close to the reference point is as follows: calculating the Euclidean distance between each non-dominated solution and the reference point, and determining non-dominated solutions with Euclidean distances less than a preset threshold as individuals with high non-dominated levels.

[0052] In one possible implementation, dynamically adjusting the displacement parameters of the model points based on the cluster analysis results of the plurality of non-dominated optimal solutions and outputting the optimal aerodynamic shape design of the object when a preset convergence condition is met includes: a: Perform cluster analysis on multiple non-dominated optimal solutions to identify solution clusters with different performance tendencies.

[0053] Here, cluster analysis is performed on the distribution characteristics of multiple non-dominated optimal solutions in the target space.

[0054] Here, “solution cluster” refers to the subset with similar performance characteristics or target preferences identified after clustering the non-dominated optimal solutions.

[0055] b: Directedly shrink the spatial range of the displacement parameters of the model points of each solution cluster. In the next generation of evolution, according to the distribution density of each solution cluster in the target space, continue to use the reference point mechanism and adaptive population allocation strategy to dynamically adjust the distribution ratio of the population among the clusters.

[0056] Here, the position and density of reference points are dynamically adjusted based on the results of cluster analysis. For example, "cold spots" (rarely occupied reference points) are deleted, "hot spots" (areas with dense solution sets) are refined, or reference points are moved around to better cover the actual Pareto front. In the next generation of the population, solutions close to sparse reference points are preferentially retained to ensure a diverse and evenly distributed solution set.

[0057] c: When the aerodynamic performance parameters of the solution cluster meet the preset convergence conditions, the solution cluster is used as the optimal aerodynamic shape design solution for the object.

[0058] Here, a new generation of population is generated through genetic algorithm operations such as crossover and mutation, and the above steps are continued. When the aerodynamic performance parameters of the solution cluster meet the preset convergence conditions, the solution cluster is used as the optimal aerodynamic shape design solution for the object.

[0059] Here, the aerodynamic performance parameters of the solution cluster are also determined by the aerodynamic performance prediction model.

[0060] In this application, multi-objective optimization can utilize methods such as NSGA-III (third-generation non-dominated sorting genetic algorithm). Using automobiles as an example, NSGA-III can be used to optimize aerodynamic shape, efficiently balancing multiple conflicting objectives (e.g., reducing drag coefficient Cd, increasing downforce Cl, and reducing aerodynamic noise). This method utilizes a reference point mechanism to process high-dimensional objective spaces, combines parametric DFFD modeling to generate diverse design samples, and drives LBM / CFD simulations to evaluate the performance of each solution. Specifically, this application designs a hierarchical contraction strategy to address the characteristics of multi-objective optimization. First, cluster analysis is performed on non-dominated optimal solutions to identify clusters of solutions with different performance trends. Then, targeted contraction is performed on the DFFD parameter subspaces corresponding to each cluster, maintaining efficient search capabilities in three-dimensional and higher-dimensional objective spaces. NSGA-III's adaptive population allocation strategy effectively maintains the distribution of the Pareto frontier. Combined with the intelligent contraction method of this invention, this strategy avoids the performance degradation of traditional multi-objective algorithms when dealing with more than three objectives, significantly improving optimization efficiency.

[0061] In one possible implementation, after outputting the aerodynamic performance parameters of each of the new three-dimensional geometric models, the aerodynamic shape optimization method further includes: If the aerodynamic performance parameter of the new three-dimensional geometric model is one, the displacement parameters of the model points corresponding to the multiple new three-dimensional geometric models are iteratively optimized based on the genetic algorithm, and the optimal aerodynamic shape design scheme of the object is output when the preset convergence condition is met.

[0062] Here, if the aerodynamic performance parameter is one, the optimization target is one. The displacement parameters of the model points corresponding to multiple new three-dimensional geometric models are iteratively optimized according to the genetic algorithm. When the preset convergence conditions are met, the optimal aerodynamic shape design scheme of the object is output.

[0063] In one possible implementation, the iterative optimization of displacement parameters of the model points corresponding to the plurality of new three-dimensional geometric models based on a genetic algorithm, and outputting an optimal aerodynamic shape design scheme for the object when a preset convergence condition is met, includes: i: Using the displacement parameters of the plurality of model points as the initial population in the genetic algorithm.

[0064] ii: Taking the aerodynamic performance parameters as the fitness function, a genetic algorithm is used to perform multi-generation iterative optimization on the population, generating a new generation of individuals in each generation, and retaining individuals with high fitness as the optimal solution.

[0065] Here, a genetic algorithm is used to perform multi-generation iterative optimization on the population to generate a new generation of individuals in each generation. The fitness of the new generation of individuals is determined according to the fitness function, and the individuals with high fitness are regarded as the optimal solution.

[0066] iii: Determine a search interval for the displacement parameters based on the distribution of the displacement parameters of the model points of the optimal solution in the current generation, and expand a buffer boundary of a preset proportion based on the search interval.

[0067] Here, the search interval of the displacement parameters is determined according to the distribution of the displacement parameters of the model points of the optimal solution in the current generation, and the buffer boundary of a preset proportion is expanded based on the search interval.

[0068] iv: When the aerodynamic performance parameters of the individual corresponding to the displacement parameters of the model points within the search interval meet the preset convergence conditions, the individual and its corresponding three-dimensional geometric model and aerodynamic performance parameters are used as the optimal aerodynamic shape design scheme.

[0069] In the specific implementation method, the aerodynamic performance parameter is the drag coefficient, the displacement of the model point is encoded as the initial population, and the drag coefficient is used as the fitness function. Through multiple generations of iteration, high-performance individuals are gradually screened and the dominant characteristics are retained as the optimal solution. This solution introduces an adaptive parameter space contraction strategy: during the optimization process, the search range is dynamically adjusted based on the parameter distribution of the optimal solution. The displacement parameters of the DFFD model points are contracted every 10 generations, and the parameter interval of the 20% of individuals with the best performance is retained. A 10% buffer boundary is given, thereby significantly improving the local search efficiency while maintaining population diversity. Compared with gradient optimization algorithms, genetic algorithms can avoid falling into local optimality and are particularly suitable for high-dimensional, nonlinear aerodynamic optimization problems with multiple extreme points.

[0070] For further information, see Figure 2 , Figure 2 This is a schematic diagram of a method for optimizing the aerodynamic shape of a complex object provided in an embodiment of the present application. Figure 2 As shown, a complex geometric basic model is selected for DFFD deformation to form a database of complex geometric shapes, and the database of complex geometric shapes is input into the aerodynamic performance prediction model in batches to obtain the aerodynamic performance parameters of the complex geometric shapes. It is detected whether the aerodynamic performance parameters meet the convergence conditions. If not, based on the intelligent optimization algorithm, different model points of complex geometric shapes are continuously found for DFFD deformation. The aerodynamic performance prediction model is used to predict the aerodynamic performance parameters of the deformed geometry. If the convergence conditions are met, the optimal aerodynamic shape design scheme is output.

[0071] This invention effectively solves the key bottlenecks of traditional aerodynamic optimization in terms of design efficiency, global search capability, degree of automation and dependence on computing resources, providing an efficient, accurate and low-cost solution for the rapid optimization of complex aerodynamic shapes, significantly improving the practicality of engineering design and its potential for industrial application.

[0072] An embodiment of the present application provides a method for optimizing the aerodynamic shape of a complex object. The method comprises: performing displacement parameter transformation on multiple model points of an initial three-dimensional geometric model of the object, then deforming the initial three-dimensional geometric model using a direct free-form deformation technique to generate multiple new three-dimensional geometric models; performing aerodynamic performance prediction on the point cloud data of the multiple new three-dimensional geometric models based on an aerodynamic performance prediction model, and outputting aerodynamic performance parameters for each of the new three-dimensional geometric models; if the new three-dimensional geometric models have multiple aerodynamic performance parameters, searching and selecting multiple non-dominated optimal solutions from the multiple new three-dimensional geometric models based on a non-dominated sorting and reference point guidance mechanism; dynamically adjusting the displacement parameters of the model points based on cluster analysis results of the multiple non-dominated optimal solutions; and outputting an optimal aerodynamic shape design solution for the object when a preset convergence condition is met. By using the direct free-form deformation technique, the aerodynamic performance prediction model, and the intelligent optimization process for multiple aerodynamic performance parameters, an optimal aerodynamic shape design solution for the object can be quickly and accurately generated.

[0073] See also Figure 3 、 Figure 4 , Figure 3 This is one of the structural schematic diagrams of a device for optimizing the aerodynamic shape of a complex object provided in an embodiment of the present application; Figure 4 This is a second structural diagram of a device for optimizing the aerodynamic shape of a complex object provided in an embodiment of the present application. Figure 3 As shown in FIG, the aerodynamic shape optimization device 300 for a complex object includes: A free deformation processing module 310 is configured to transform the displacement parameters of multiple model points of the initial 3D geometric model of the object based on a direct free deformation technique, and then deform the initial 3D geometric model using the direct free deformation technique to generate multiple new 3D geometric models. an aerodynamic performance prediction module 320 for performing aerodynamic performance prediction processing on the point cloud data of the plurality of new three-dimensional geometric models based on an aerodynamic performance prediction model, and outputting aerodynamic performance parameters of each of the new three-dimensional geometric models; The intelligent optimization module 330 is used to search and select multiple non-dominated optimal solutions from the multiple new three-dimensional geometric models based on the non-dominated sorting and reference point guidance mechanism if there are multiple aerodynamic performance parameters of the new three-dimensional geometric model, dynamically adjust the displacement parameters of the model points based on the clustering analysis results of the multiple non-dominated optimal solutions, and output the optimal aerodynamic shape design scheme of the object when the preset convergence conditions are met.

[0074] Furthermore, the intelligent optimization module 330 is further configured to: If the aerodynamic performance parameter of the new three-dimensional geometric model is one, the displacement parameters of the model points corresponding to the multiple new three-dimensional geometric models are iteratively optimized based on the genetic algorithm, and the optimal aerodynamic shape design scheme of the object is output when the preset convergence condition is met.

[0075] Furthermore, the aerodynamic performance prediction module 320 is configured to perform aerodynamic performance prediction processing on the point cloud data of a plurality of new three-dimensional geometric models based on the aerodynamic performance prediction model for any of the new three-dimensional geometric models, and output aerodynamic performance parameters of each of the new three-dimensional geometric models: Performing feature extraction and encoding processing on the new three-dimensional geometric model in the aerodynamic performance prediction model to determine voxel features corresponding to each point cloud data; Processing the plurality of voxel features based on a convolutional layer, a linear layer, a normalization layer, and an activation function, and outputting a plurality of embedded features; Processing the plurality of embedded features in sequence based on a local attention layer, a feedforward network layer, a residual connection layer, and a normalization layer in a multi-layer encoder to output a plurality of multi-scale features; Global pooling is performed on the plurality of multi-scale features to output global features, and mapping is performed on the global features based on a regressor to output the aerodynamic performance parameters.

[0076] Furthermore, the intelligent optimization module 330 is configured to search and select multiple non-dominated optimal solutions in the multiple new three-dimensional geometric models based on the non-dominated sorting and reference point guidance mechanism, including: A population is formed based on the multiple new three-dimensional geometric models and the corresponding aerodynamic performance parameters, and all individuals in the population are sorted according to non-domination relationships to generate a non-dominated hierarchy; wherein the first level of the non-dominated hierarchy is a completely non-dominated solution set, and the second level is a non-dominated solution set remaining after removing the first level; A set of reference points are evenly arranged in the target optimization space, and when selecting the next generation of individuals from the population, individuals with high non-dominated levels close to the reference points are preferentially selected, and the individuals with high non-dominated levels are used as the non-dominated optimal solutions; wherein the reference points are optimization targets corresponding to multiple aerodynamic performance parameters.

[0077] Furthermore, the intelligent optimization module 330 is used to dynamically adjust the displacement parameters of the model points based on the cluster analysis results of the plurality of non-dominated optimal solutions, and output the optimal aerodynamic shape design solution of the object when a preset convergence condition is met: performing cluster analysis on the plurality of non-dominated optimal solutions to identify solution clusters with different performance tendencies; Directedly shrink the spatial range of the displacement parameters of the model points of each solution cluster. In the next generation of evolution, according to the distribution density of each solution cluster in the target space, continue to use the reference point mechanism and adaptive population allocation strategy to dynamically adjust the distribution ratio of the population among the clusters. When the aerodynamic performance parameters of the solution cluster meet the preset convergence conditions, the solution cluster is used as the optimal aerodynamic shape design solution for the object.

[0078] Furthermore, the intelligent optimization module 330 is used to iteratively optimize the displacement parameters of the model points corresponding to the multiple new three-dimensional geometric models based on the genetic algorithm, and output the optimal aerodynamic shape design scheme of the object when the preset convergence conditions are met: Using the displacement parameters of the plurality of model points as the initial population in the genetic algorithm; Taking the aerodynamic performance parameters as fitness functions, using a genetic algorithm to perform multi-generation iterative optimization on the population, generating a new generation of individuals in each generation, and retaining individuals with high fitness as the optimal solution; Determine a search interval for the displacement parameters based on the distribution of the displacement parameters of the model points of the optimal solution in the current generation, and expand a buffer boundary of a preset proportion based on the search interval; When the individual aerodynamic performance parameters corresponding to the displacement parameters of the model points within the search interval meet the preset convergence conditions, the individual and its corresponding three-dimensional geometric model and aerodynamic performance parameters are used as the optimal aerodynamic shape design scheme.

[0079] Further, such as Figure 4 As shown, the aerodynamic shape optimization device 300 for a complex object further includes a model training module 340. The model training module 340 determines the aerodynamic performance prediction model through the following steps: Inputting sample point cloud data of multiple sample three-dimensional geometric models and corresponding sample aerodynamic performance parameters into a deep learning network model to learn the mapping relationship between the sample point cloud data and the sample aerodynamic performance parameters, and predicting the predicted aerodynamic performance parameters of the sample point cloud data; The deep learning network model is iteratively trained based on the loss value between the predicted aerodynamic performance parameters and the sample aerodynamic performance parameters to determine the aerodynamic performance prediction model.

[0080] An embodiment of the present application provides an aerodynamic shape optimization device for a complex object. The device comprises: a free deformation processing module for performing displacement parameter transformation on multiple model points of an initial three-dimensional geometric model of the object based on a direct free deformation technique, and then deforming the initial three-dimensional geometric model using the direct free deformation technique to generate multiple new three-dimensional geometric models; an aerodynamic performance prediction module for performing aerodynamic performance prediction on point cloud data of the multiple new three-dimensional geometric models based on the aerodynamic performance prediction model, and outputting aerodynamic performance parameters for each of the new three-dimensional geometric models; and an intelligent optimization module for selecting multiple non-dominated optimal solutions from the multiple new three-dimensional geometric models based on a non-dominated sorting and reference point guidance mechanism, dynamically adjusting the displacement parameters of the model points based on cluster analysis results of the multiple non-dominated optimal solutions, and outputting an optimal aerodynamic shape design solution for the object when a preset convergence condition is met. By utilizing the direct free deformation technique, the aerodynamic performance prediction model, and the intelligent optimization process for multiple aerodynamic performance parameters, an optimal aerodynamic shape design solution for the object can be quickly and accurately generated.

[0081] See also Figure 5 , Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown in FIG, the electronic device 500 includes a processor 510, a memory 520 and a bus 530.

[0082] The memory 520 stores machine-readable instructions executable by the processor 510. When the electronic device 500 is running, the processor 510 communicates with the memory 520 via the bus 530. When the machine-readable instructions are executed by the processor 510, the above-mentioned Figure 1 as well as Figure 2 The steps of the method for optimizing the aerodynamic shape of a complex object in the method embodiment shown are specifically implemented in accordance with the method embodiment and will not be described in detail here.

[0083] The embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the computer program can execute the above-mentioned Figure 1 as well as Figure 2 The steps of the method for optimizing the aerodynamic shape of a complex object in the method embodiment shown are specifically implemented in accordance with the method embodiment and will not be described in detail here.

[0084] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0085] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. There may be other division methods in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some communication interface, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0086] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0087] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0088] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0089] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The scope of protection of the present application is not limited thereto. Although the present application has been described in detail with reference to the above-mentioned embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above-mentioned embodiments within the technical scope disclosed in the present application, or perform equivalent replacements for some of the technical features thereof. These modifications, changes, or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for optimizing the aerodynamic shape of a complex object, characterized in that: The aerodynamic shape optimization method comprises: After performing displacement parameter transformation on multiple model points of the initial three-dimensional geometric model of the object, the initial three-dimensional geometric model is deformed by combining direct free deformation technology to generate multiple new three-dimensional geometric models; performing aerodynamic performance prediction processing on the point cloud data of the plurality of new three-dimensional geometric models based on the aerodynamic performance prediction model, and outputting aerodynamic performance parameters of each of the new three-dimensional geometric models; If there are multiple aerodynamic performance parameters for the new three-dimensional geometric model, multiple non-dominated optimal solutions are searched and selected from the multiple new three-dimensional geometric models based on the non-dominated sorting and reference point guidance mechanism. The displacement parameters of the model points are dynamically adjusted based on the clustering analysis results of the multiple non-dominated optimal solutions. When the preset convergence conditions are met, the optimal aerodynamic shape design scheme of the object is output.

2. The aerodynamic shape optimization method according to claim 1, characterized in that: After outputting the aerodynamic performance parameters of each of the new three-dimensional geometric models, the aerodynamic shape optimization method further includes: If the aerodynamic performance parameter of the new three-dimensional geometric model is one, the displacement parameters of the model points corresponding to the multiple new three-dimensional geometric models are iteratively optimized based on the genetic algorithm, and the optimal aerodynamic shape design scheme of the object is output when the preset convergence condition is met.

3. The aerodynamic shape optimization method according to claim 1, characterized in that: For any of the new three-dimensional geometric models, performing aerodynamic performance prediction processing on the point cloud data of the multiple new three-dimensional geometric models based on the aerodynamic performance prediction model and outputting aerodynamic performance parameters of each of the new three-dimensional geometric models includes: Performing feature extraction and encoding processing on the new three-dimensional geometric model in the aerodynamic performance prediction model to determine voxel features corresponding to each point cloud data; Processing the plurality of voxel features based on a convolutional layer, a linear layer, a normalization layer, and an activation function, and outputting a plurality of embedded features; Processing the plurality of embedded features in sequence based on a local attention layer, a feedforward network layer, a residual connection layer, and a normalization layer in a multi-layer encoder to output a plurality of multi-scale features; Global pooling is performed on the plurality of multi-scale features to output global features, and mapping is performed on the global features based on a regressor to output the aerodynamic performance parameters.

4. The aerodynamic shape optimization method according to claim 1, characterized in that: The method of searching and selecting a plurality of non-dominated optimal solutions in the plurality of new three-dimensional geometric models based on the non-dominated sorting and reference point guidance mechanism includes: A population is formed based on the multiple new three-dimensional geometric models and the corresponding aerodynamic performance parameters, and all individuals in the population are sorted according to non-domination relationships to generate a non-dominated hierarchy; wherein the first level of the non-dominated hierarchy is a completely non-dominated solution set, and the second level is a non-dominated solution set remaining after removing the first level; A set of reference points are evenly arranged in the target optimization space, and when selecting the next generation of individuals from the population, individuals with high non-dominated levels close to the reference points are preferentially selected, and the individuals with high non-dominated levels are used as the non-dominated optimal solutions; wherein the reference points are optimization targets corresponding to multiple aerodynamic performance parameters.

5. The aerodynamic shape optimization method according to claim 1, characterized in that: The method dynamically adjusts the displacement parameters of the model points based on the cluster analysis results of the plurality of non-dominated optimal solutions, and outputs the optimal aerodynamic shape design scheme of the object when a preset convergence condition is met, including: performing cluster analysis on the plurality of non-dominated optimal solutions to identify solution clusters with different performance tendencies; Directedly shrink the spatial range of the displacement parameters of the model points of each solution cluster. In the next generation of evolution, according to the distribution density of each solution cluster in the target space, continue to use the reference point mechanism and adaptive population allocation strategy to dynamically adjust the distribution ratio of the population among the clusters. When the aerodynamic performance parameters of the solution cluster meet the preset convergence conditions, the solution cluster is used as the optimal aerodynamic shape design solution for the object.

6. The aerodynamic shape optimization method according to claim 2, characterized in that: The iterative optimization of displacement parameters of model points corresponding to the plurality of new three-dimensional geometric models based on a genetic algorithm, and outputting an optimal aerodynamic shape design scheme for the object when a preset convergence condition is met, comprises: Using the displacement parameters of the plurality of model points as the initial population in the genetic algorithm; Taking the aerodynamic performance parameters as fitness functions, using a genetic algorithm to perform multi-generation iterative optimization on the population, generating a new generation of individuals in each generation, and retaining individuals with high fitness as the optimal solution; Determine a search interval for the displacement parameters based on the distribution of the displacement parameters of the model points of the optimal solution in the current generation, and expand a buffer boundary of a preset proportion based on the search interval; When the individual aerodynamic performance parameters corresponding to the displacement parameters of the model points within the search interval meet the preset convergence conditions, the individual and its corresponding three-dimensional geometric model and aerodynamic performance parameters are used as the optimal aerodynamic shape design scheme.

7. The aerodynamic shape optimization method according to claim 1, characterized in that: The aerodynamic performance prediction model is determined by the following steps: Inputting sample point cloud data of multiple sample three-dimensional geometric models and corresponding sample aerodynamic performance parameters into a deep learning network model to learn the mapping relationship between the sample point cloud data and the sample aerodynamic performance parameters, and predicting the predicted aerodynamic performance parameters of the sample point cloud data; The deep learning network model is iteratively trained based on the loss value between the predicted aerodynamic performance parameters and the sample aerodynamic performance parameters to determine the aerodynamic performance prediction model.

8. An aerodynamic shape optimization device for a complex object, characterized in that: The aerodynamic shape optimization device comprises: A free deformation processing module is used to transform the displacement parameters of multiple model points of the initial three-dimensional geometric model of the object, and then deform the initial three-dimensional geometric model using direct free deformation technology to generate multiple new three-dimensional geometric models; an aerodynamic performance prediction module, configured to perform aerodynamic performance prediction processing on the point cloud data of the plurality of new three-dimensional geometric models based on an aerodynamic performance prediction model, and output aerodynamic performance parameters of each of the new three-dimensional geometric models; An intelligent optimization module is used to search and select multiple non-dominated optimal solutions from the multiple new three-dimensional geometric models based on a non-dominated sorting and reference point guidance mechanism if there are multiple aerodynamic performance parameters of the new three-dimensional geometric model, dynamically adjust the displacement parameters of the model points based on the clustering analysis results of the multiple non-dominated optimal solutions, and output the optimal aerodynamic shape design scheme of the object when the preset convergence conditions are met.

9. An electronic device, characterized in that: include: A processor, a memory and a bus, wherein the memory stores machine-readable instructions executable by the processor. When the electronic device is running, the processor and the memory communicate through the bus. When the processor is running, the machine-readable instructions execute the steps of the aerodynamic shape optimization method for a complex object as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, which, when executed by a processor, executes the steps of the method for optimizing the aerodynamic shape of a complex object as claimed in any one of claims 1 to 7.

Images