Axial flow compressor airfoil optimization method based on primary and secondary agent model and related device

Abstract

The application discloses an axial-flow compressor blade profile optimization method based on a master-slave agent model and related devices, and relates to the field of compressor blade design. The method comprises the following steps: generating a plurality of design parameter groups based on to-be-optimized design parameters and parameter adjustable ranges of to-be-optimized axial-flow compressor blades by using a data sampling algorithm, outputting respective corresponding blade profile performance prediction values of the design parameter groups by using a master-slave combined agent model, screening a plurality of alternative design parameter groups based on a comparison result of the respective blade profile performance prediction values and an initial performance benchmark, performing computational fluid dynamics simulation, and determining an alternative design parameter group corresponding to optimal simulation performance parameters as a blade profile optimization parameter group. The application reduces the invalid risk of the design parameter group by using the parameter adjustable range, improves the performance prediction accuracy based on flow field characteristic parameters by using the master-slave combined agent model, and finally determines the blade profile optimization parameter group based on the simulation result of the screened alternative design parameter group, thereby further improving the prediction accuracy.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of compressor blade design, and particularly relates to an axial flow compressor blade profile optimization method based on a master-slave surrogate model and a related device. BACKGROUND

[0002] An axial flow compressor is a key device in the fields of aerospace and energy, and the blade profile design directly determines the aerodynamic performance. Traditional design relies on the experience of engineers to manually adjust parameters, which is time-consuming and prone to local optimization. To solve the problem of high computational cost of existing technologies based on optimization algorithms and computational fluid dynamics (CFD) simulation, a surrogate model is constructed to replace high-cost CFD analysis.

[0003] However, since the existing surrogate model only focuses on the mapping of design variables and performance output, a large amount of flow field data generated by CFD simulation is wasted, resulting in the problem of insufficient prediction accuracy of the surrogate model trained under a small sample. Therefore, how to improve the prediction accuracy of the surrogate model trained under a small sample for blade optimization parameters has become a problem to be solved. SUMMARY

[0004] In view of the above problems, the present application provides an axial flow compressor blade profile optimization method based on a master-slave surrogate model and a related device to improve the prediction accuracy of the surrogate model trained under a small sample for blade optimization parameters. The specific scheme is as follows:

[0005] The first aspect of the present application provides an axial flow compressor blade profile optimization method based on a master-slave surrogate model, comprising:

[0006] obtaining a to-be-optimized design parameter of a to-be-optimized axial flow compressor blade and an initial performance benchmark, the to-be-optimized design parameter being a blade adjustable geometric parameter affecting the aerodynamic performance of the blade profile, and the initial performance benchmark being a result parameter obtained based on computational fluid dynamics simulation of an original blade profile of the to-be-optimized axial flow compressor blade;

[0007] generating a plurality of design parameter groups based on the to-be-optimized design parameter and a parameter adjustable range by using a data sampling algorithm;

[0008] inputting each of the design parameter groups into a master-slave combined surrogate model in sequence to obtain a blade profile performance prediction value corresponding to each of the design parameter groups, and the training data of the master-slave combined surrogate model at least including a flow field characteristic parameter obtained based on computational fluid dynamics simulation of the original blade profile;

[0009] The multi-objective optimization algorithm is used to compare the blade profile performance prediction values with the initial performance benchmark, and based on the comparison result, a plurality of candidate design parameter groups are selected from each of the design parameter groups;

[0010] Based on each of the candidate design parameter groups, a computational fluid dynamics simulation is performed to obtain simulation performance parameters of each of the candidate design parameter groups, and the candidate design parameter group corresponding to the optimal simulation performance parameter is determined as the blade profile optimization parameter group.

[0011] In one possible implementation, the training process of the main-vice combined proxy model includes:

[0012] Obtain the sweep parameters of a plurality of groups of control points of the original blade profile, and the sweep parameters of a group of control points include the sweep values of each control point of the Bezier curve of the original blade profile of the axial flow compressor blade to be optimized;

[0013] Based on the sweep parameters of each group of control points, a computational fluid dynamics simulation is performed to obtain the stable operating range parameters and fluid characteristic parameters corresponding to each group of sweep parameters of the control points;

[0014] A data set including each group of sweep parameters of the control points added with a label is determined as a first training set, and the content of the label is the stable operating range parameters corresponding to the sweep parameters of the control points; and a data set including each of the fluid characteristic parameters is determined as a second training set;

[0015] Based on the first training set and the second training set, an initial main-vice proxy model is trained to obtain the main-vice combined proxy model, and the input of the main-vice combined proxy model is a plurality of design parameter groups generated based on the to-be-optimized design parameters and the parameter adjustable range of the axial flow compressor blade to be optimized, and the output is the blade profile performance prediction value corresponding to each design parameter group.

[0016] In one possible implementation, the training, based on the first training set and the second training set, of the initial main-vice proxy model to obtain the main-vice combined proxy model includes,

[0017] The initial main-vice proxy model is initialized, and the main-vice proxy model includes a main model, a vice model, a fusion layer, and a linear transformation layer;

[0018] The first training set is input into the main model to obtain initial blade profile performance prediction results of the sweep parameters of each group of control points output by the main model;

[0019] The second training set is input into the vice model to obtain auxiliary blade profile performance prediction results of the sweep parameters of each group of control points output by the vice model;

[0020] The initial blade profile performance prediction results and the auxiliary blade profile performance prediction results are input into the fusion layer, so that the fusion layer sums the initial blade profile performance prediction results and the auxiliary blade profile performance prediction results corresponding to the sweep parameters of each group of control points, adjusts the sum weight by calling the linear transformation layer, obtains the blade profile performance prediction values of the sweep parameters of each group of control points, and adjusts the parameters of the primary model and the secondary model based on the blade profile performance prediction values and the initial performance benchmark, to obtain the primary-secondary combined proxy model.

[0021] In a possible implementation, the blade profile performance prediction values at least include stable working prediction ranges and peak efficiency prediction working conditions, the initial performance benchmark at least includes initial stable working ranges and initial peak efficiency working conditions, and the multi-objective optimization algorithm is used to compare each blade profile performance prediction value with the initial performance benchmark, and based on the comparison result, a plurality of alternative design parameter groups are screened from each design parameter group, including:

[0022] The multi-objective optimization algorithm is used to compare each stable working prediction range with the initial stable working range, compare each peak efficiency prediction working condition with the initial peak efficiency working condition, and determine the design parameter group corresponding to the blade profile performance prediction value that is not less than the initial stable working range in the stable working prediction range and not less than the initial peak efficiency working condition in the peak efficiency prediction working condition as the alternative design parameter group.

[0023] In a possible implementation, the type of the fluid characteristic parameter is at least one or more of a high-entropy region value, a maximum efficiency point, an axial momentum of the to-be-optimized axial flow compressor blade, and a position of a center of aerodynamic load.

[0024] In a possible implementation, the method further includes:

[0025] In the absence of the alternative design parameter group, the design parameter group corresponding to each blade profile performance prediction value is used as supplementary training data, and the primary-secondary combined proxy model is optimized and trained by using the supplementary training data;

[0026] In the case where the primary-secondary combined proxy model is completed optimization training, the operation step of sequentially inputting each design parameter group into the primary-secondary combined proxy model to obtain the blade profile performance prediction value corresponding to each design parameter group is performed.

[0027] The second aspect of the present application provides an axial flow compressor blade profile optimization system based on a primary-secondary proxy model, including:

[0028] a parameter obtaining module configured to obtain an initial performance benchmark and an optimization design parameter of an optimization axial compressor blade, the optimization design parameter being an adjustable geometry parameter of the blade affecting aerodynamic performance of a blade profile, and the initial performance benchmark being a result parameter obtained by performing computational fluid dynamics simulation based on an original blade profile of the optimization axial compressor blade;

[0029] a parameter generating module configured to generate a plurality of design parameter groups based on the optimization design parameter and a parameter adjustable range by using a data sampling algorithm;

[0030] a performance predicting module configured to input each of the design parameter groups into a main-sub combined proxy model in sequence to obtain a blade profile performance prediction value corresponding to each of the design parameter groups, the training data of the main-sub combined proxy model including at least a flow field characteristic parameter obtained by performing computational fluid dynamics simulation based on the original blade profile;

[0031] a performance comparing module configured to compare each of the blade profile performance prediction values with the initial performance benchmark by using a multi-objective optimization algorithm, and to screen a plurality of candidate design parameter groups from the design parameter groups based on a comparison result;

[0032] a parameter outputting module configured to perform computational fluid dynamics simulation based on each of the candidate design parameter groups to obtain a simulation performance parameter of each of the candidate design parameter groups, and to determine a candidate design parameter group corresponding to an optimal simulation performance parameter as a blade profile optimization parameter group.

[0033] In a possible implementation, the optimization system of the axial compressor blade further includes a model training module configured to:

[0034] obtain a plurality of groups of sweep parameters of control points, a sweep parameter of a group of the control points including a sweep value of each control point of a Bezier curve of the optimization axial compressor blade of the original blade profile;

[0035] perform computational fluid dynamics simulation based on each group of the sweep parameters of the control points to obtain a stable working range parameter and a fluid characteristic parameter corresponding to each group of the sweep parameters of the control points;

[0036] determine a data set including each group of the sweep parameters of the control points added with a label as a first training set, a content of the label being the stable working range parameter corresponding to the sweep parameter of the control point, and determine a data set including each of the fluid characteristic parameters as a second training set;

[0037] training an initial master-slave agent model based on the first training set and the second training set to obtain the master-slave combined agent model, wherein an input of the master-slave combined agent model is a plurality of design parameter groups generated based on to-be-optimized design parameters and parameter adjustable ranges of a to-be-optimized axial flow compressor blade, and an output is a respective corresponding blade profile performance prediction value of each design parameter group.

[0038] In a possible implementation, the model training module is configured to:

[0039] initialize the initial master-slave agent model, wherein the master-slave agent model comprises a master model, a slave model, a fusion layer, and a linear transformation layer;

[0040] input the first training set into the master model to obtain initial blade profile performance prediction results of the control point of each group output by the master model;

[0041] input the second training set into the slave model to obtain auxiliary blade profile performance prediction results of the control point of each group output by the slave model;

[0042] input the initial blade profile performance prediction results and the auxiliary blade profile performance prediction results into the fusion layer, so that the fusion layer sums the initial blade profile performance prediction results and the auxiliary blade profile performance prediction results corresponding to the control point of each group, calls the linear transformation layer to adjust the sum weight, obtains blade profile performance prediction values of the control point of each group, and adjusts parameters of the master model and the slave model based on the blade profile performance prediction values and the initial performance benchmark to obtain the master-slave combined agent model.

[0043] In a possible implementation, the performance comparison module is configured to:

[0044] compare each stable working prediction range with an initial stable working range and each peak efficiency prediction working condition with an initial peak efficiency working condition by using the multi-objective optimization algorithm, and determine the design parameter group corresponding to the blade profile performance prediction value, which is greater than or equal to the initial stable working range and the initial peak efficiency working condition, as the candidate design parameter group, wherein the blade profile performance prediction value at least includes the stable working prediction range and the peak efficiency prediction working condition, and the initial performance benchmark at least includes the initial stable working range and the initial peak efficiency working condition.

[0045] In a possible implementation, the type of the fluid characteristic parameter is at least one or more of a high-entropy region value, a maximum efficiency point, an axial momentum of the axial-flow compressor blade to be optimized, and a position of an aerodynamic load center.

[0046] In a possible implementation, the performance comparison module is further configured to:

[0047] In the absence of the set of alternative design parameters, the set of design parameters corresponding to each of the blade performance prediction values is used as supplementary training data, and the main-aided combined surrogate model is optimized and trained using the supplementary training data.

[0048] In the case where the optimization and training of the main-aided combined surrogate model is completed, the performance prediction module is triggered to perform the operation of sequentially inputting each of the sets of design parameters into the main-aided combined surrogate model to obtain the blade performance prediction value corresponding to each of the sets of design parameters.

[0049] The third aspect of the present application provides a computer program product, which includes computer readable instructions, and when the computer readable instructions are run on an electronic device, the electronic device implements the main-aided combined surrogate model-based axial-flow compressor blade optimization method of the first aspect or any implementation manner of the first aspect.

[0050] The fourth aspect of the present application provides an electronic device, which includes at least one processor and a memory connected to the processor, and wherein:

[0051] The memory is configured to store a computer program.

[0052] The processor is configured to execute the computer program, so that the electronic device can implement the main-aided combined surrogate model-based axial-flow compressor blade optimization method of the first aspect or any implementation manner of the first aspect.

[0053] The fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device can implement the main-aided combined surrogate model-based axial-flow compressor blade optimization method of the first aspect or any implementation manner of the first aspect.

[0054] By the technical scheme, the axial-flow compressor blade profile optimization method and related device based on the master-slave agent model are provided, the selected optimization design parameters affecting the aerodynamic performance of the blade profile of the axial-flow compressor to be optimized are obtained by configuration, and the initial performance benchmark obtained by the computational fluid dynamics simulation based on the original blade profile of the axial-flow compressor blade to be optimized, so that the optimization object and the reference standard are determined. Subsequently, a plurality of design parameter groups are generated based on the optimization design parameters and the parameter adjustable range by using the data sampling algorithm, so that the parameter invalidation risk in the design parameter group is reduced by using the parameter adjustable range, and the prediction and optimization accuracy in the subsequent process is improved. Moreover, the training data of the master-slave combined agent model at least includes the flow field characteristic parameters obtained by the computational fluid dynamics simulation based on the original blade profile, so that the master-slave combined agent model can predict the performance of the design parameter group based on the flow field characteristics captured in the training process. Subsequently, the multi-objective optimization algorithm is used to compare the blade profile performance prediction value with the initial performance benchmark, and a plurality of alternative design parameter groups are selected from the design parameter groups based on the comparison result, so as to reduce the calculation amount of the subsequent computational fluid dynamics simulation, and improve the simulation accuracy of the simulation performance parameters obtained subsequently. Finally, the computational fluid dynamics simulation is performed based on the alternative design parameter groups, the simulation performance parameters of the alternative design parameter groups are obtained, and the alternative design parameter group corresponding to the optimal simulation performance parameter is determined as the blade profile optimization parameter group, so as to improve the prediction accuracy of the finally determined blade profile optimization parameter group. BRIEF DESCRIPTION OF DRAWINGS

[0055] The above and other features, advantages, and aspects of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the drawings, the same or similar reference numerals are used to denote the same or similar elements. It is to be understood that the drawings are schematic, and the proportions of the elements and the like are not necessarily to scale.

[0056] Figure 1 A flowchart of the axial-flow compressor blade profile optimization method based on the master-slave agent model provided by the present application;

[0057] Figure 2 A schematic diagram of a Bezier curve provided by the present application;

[0058] Figure 3 An architecture schematic diagram of an initial master-slave agent model provided by the present application;

[0059] Figure 4 A schematic diagram of a high-entropy region of a blade provided by the present application;

[0060] Figure 5 A flow-rate-pressure ratio characteristic curve diagram provided by the present application;

[0061] Figure 6A flow-isentropic efficiency characteristic curve diagram provided for the present application;

[0062] Figure 7 A block diagram of an axial flow compressor blade profile optimization system provided for the present application;

[0063] Figure 8 A structural schematic diagram of an electronic device provided for the present application. DETAILED DESCRIPTION

[0064] The embodiments of the present application are described below in conjunction with the accompanying drawings. The terms used in the embodiment part of the present application are only used to explain the specific embodiments of the present application, and are not intended to limit the present application.

[0065] The embodiments of the present application are described below in conjunction with the accompanying drawings. It is known to those skilled in the art that with the development of technology and the appearance of new scenarios, the technical solutions provided by the embodiments of the present application are also applicable to similar technical problems.

[0066] The terms "first", "second", and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily mean a specific order or sequence. It should be understood that the terms used in this way can be interchanged under appropriate circumstances, and this is only a distinguishing way used in the description of the embodiments of the present application to describe the objects with the same attributes. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, so that the processes, methods, systems, products or equipment containing a series of units do not have to be limited to those units, but can include other units not clearly listed or inherent to these processes, methods, products or equipment.

[0067] The first aspect of the present application provides an axial flow compressor blade profile optimization method based on a master-slave agent model, as shown in the figure, the axial flow compressor blade profile optimization method based on the master-slave agent model includes: Figure 1

[0068] S101, obtaining the initial performance benchmark and the to-be-optimized design parameter of the to-be-optimized axial flow compressor blade, the to-be-optimized design parameter is the blade adjustable geometric parameter affecting the aerodynamic performance of the blade, and the initial performance benchmark is a result parameter obtained by performing computational fluid dynamics simulation based on the original blade profile of the to-be-optimized axial flow compressor blade.

[0069] ​It should be noted that in an actual application scenario, the above blade adjustable geometry parameter is the sweep value of each control point on the Bézier curve of the to-be-optimized axial flow compressor blade, wherein the sweep value refers to the inclination angle of the blade along the radial direction, and the sweep value being positive indicates that the blade is inclined to the gas outlet along the axial direction, and the sweep value being negative indicates that the blade is inclined to the gas outlet in the opposite direction along the axial direction. Since the shape of the to-be-optimized axial flow compressor blade can be set to a regular shape (such as a rectangle) or an irregular shape (such as an arc surface or a curved surface based on its design requirements, when the shape of the blade is irregular, the sweep value of a single point is not enough to represent the aerodynamic performance of the entire blade profile. Therefore, by equally dividing the Bézier curve of the to-be-optimized axial flow compressor blade to obtain a plurality of control points, and extracting the sweep values of the control points, the accuracy of the subsequent prediction results is improved.

[0070] In a possible implementation, there are various ways to obtain the Bézier curve of the to-be-optimized axial flow compressor blade and the control points thereon, and one of them is provided as an example:

[0071] As shown in Figure 2 , it is a schematic diagram of the Bézier curve of the to-be-optimized axial flow compressor blade. The way to obtain the Bézier curve can include the following steps A1 to A3.

[0072] Step A1, divide the to-be-optimized axial flow compressor blade into 10 sections along the radial direction. And trigger step A2.

[0073] In a possible implementation, the 10 sections in step A1 are the sections designated by reference numerals 1 to 10 in Figure 2 . Among them, section 1 is the section at the blade root, and section 10 is the section at the blade tip. The direction from the center of section 1 to the center of section 10 is the radial direction in step A1.

[0074] In a possible implementation, step A1 can be realized by the "radial plane cutting" function in the modeling tool. The modeling tool includes but is not limited to: Computer Aided Design (CAD) tool, Unigraphics (UG) tool, SolidWorks, etc.

[0075] Step A2, find the geometric center (or centroid / gravity center) of each section in step A1, and determine the curve fitting each geometric center (or centroid / gravity center) as the Bézier curve of the to-be-optimized axial flow compressor blade. And trigger step A3.

[0076] In a possible implementation, the Bézier curve is shown as a broken line segment passing through each section in Figure 2 .

[0077] Step A3, equally divide the Bezier curve, and determine each division point as a control point of the Bezier curve.

[0078] In a possible implementation, each control point of the Bezier curve is determined as shown in each inflection point on the Bezier curve. Figure 2

[0079] It should be noted that in an actual application scenario, the initial performance standard based on the original blade profile of the to-be-optimized axial flow compressor blade obtained by the computational fluid dynamics simulation is obtained through configuration, the relatively true and accurate performance standard parameter is obtained, and then the subsequent prediction result is checked through the initial performance standard, the risk that the output prediction result is lower than the original blade profile performance of the to-be-optimized axial flow compressor blade is avoided, and the precision of the blade profile optimization parameter group obtained by the final prediction is improved.

[0080] In a possible implementation, the initial performance standard can include entropy efficiency parameters (for example, stable working range increase amplitude and peak efficiency) corresponding to each control point in the process of performing the computational fluid dynamics simulation on the to-be-optimized axial flow compressor blade. Wherein, the stable working range increase amplitude I can be calculated by the following formula: . is the stable working range of the optimized blade profile, is the stable working range of the original blade profile.

[0081] S102, generating a plurality of design parameter groups based on the to-be-optimized design parameters and the parameter adjustable range by using a data sampling algorithm.

[0082] It should be noted that in an actual application scenario, the type of the data sampling algorithm can be various, including but not limited to: Latin Hypercube Sampling (LHS) algorithm, Monte Carlo Sampling (MC) algorithm, Uniform Sampling (US) algorithm, etc. The type of the data sampling algorithm is not limited and described in detail.

[0083] It should be noted that in an actual application scenario, the parameter adjustable range is the adjustment range of the blade adjustable geometric parameter without causing the blade performance to be lower than the initial performance standard. The parameter adjustable range can be obtained by calibrating the to-be-optimized axial flow compressor blade. For example, the parameter adjustable range can be-5° to +5°. It should be noted that for the to-be-optimized axial flow compressor blades with the same or similar shape and inconsistent size, the same parameter adjustable range can be used, or independent calibration can be performed.

[0084] ​It should be noted that in the actual application scenario, the above design parameter group can include the random sweep value of each control point in the to-be-optimized design parameter. Wherein, the above random sweep value is within the above parameter adjustable range.

[0085] It should be noted that the present application generates a plurality of design parameter groups based on the to-be-optimized design parameters and the parameter adjustable range by configuring the data sampling algorithm, and uses the parameter adjustable range to constrain the parameters in the design parameter group, thereby avoiding the risk of obtaining parameters that cause further decline in blade performance in the obtained design parameter group, and further improving the accuracy of the subsequent predicted blade optimization parameter group.

[0086] S103, sequentially input each design parameter group into the main and auxiliary combined proxy model to obtain the blade performance prediction value corresponding to each design parameter group respectively, and the training data of the main and auxiliary combined proxy model at least includes the flow field characteristic parameters obtained by performing computational fluid dynamics simulation based on the original blade.

[0087] It should be noted that in the actual application scenario, the above flow field characteristic parameters are core indicators for quantitatively describing the state of airflow flowing through the blade, the flow rule of the airflow and the energy conversion process under the influence of the blade, and the parameters therein represent the aerodynamic performance of the blade. Since the existing proxy model only focuses on the mapping of design variables and performance output, the existing proxy model cannot capture the correlation between the flow field characteristics and the aerodynamic performance of the blade, so that in the small sample training scenario, the existing proxy model cannot fully capture the correlation characteristics of the aerodynamic performance of the blade, and the prediction accuracy of the proxy model for the blade optimization parameters is reduced. Therefore, the present application trains the main and auxiliary combined proxy model by configuring the training data at least including the flow field characteristic parameters obtained by performing computational fluid dynamics simulation based on the original blade, and outputs the blade performance prediction value corresponding to each design parameter group based on the input of each design parameter group, so that the main and auxiliary combined proxy model can predict the performance of the design parameter group based on the correlation between the flow field characteristics and the aerodynamic performance of the blade captured in the training process, and improve the accuracy of the output blade performance prediction value.

[0088] S104, comparing each blade performance prediction value with the initial performance benchmark by using a multi-objective optimization algorithm, and selecting a plurality of candidate design parameter groups from each design parameter group based on the comparison result.

[0089] It should be noted that in the actual application scenario, since any prediction model has the risk of prediction accuracy decline due to model overfitting in the prediction process. Therefore, the present application compares each blade performance prediction value with the initial performance benchmark by configuring the multi-objective optimization algorithm, and selects a plurality of candidate design parameter groups from each design parameter group based on the comparison result, realizes the elimination of the design parameter group with abnormality, and further improves the accuracy of the finally determined blade optimization parameter group.

[0090] It should be noted that in actual application scenarios, the above multi-objective optimization algorithm (Multi Objective Optimization Algorithm, MOA) is an algorithm for finding the optimal solution when considering multiple conflicting objectives. There are many types of multi-objective optimization algorithms, including but not limited to: non-dominated sorting genetic algorithm II (Non-dominated Sorting Genetic Algorithm II, NSGA-II), multi-objective evolutionary algorithm based on decomposition (Multi-Objective Evolutionary Algorithm based on Decomposition, MOEA / D), strength pareto evolutionary algorithm 2 (Strength Pareto Evolutionary Algorithm 2, SPEA2), etc. Since the above blade performance prediction value and initial performance benchmark both have multiple types of data, if only a single data is used for screening, there is a risk that the optimal blade optimization parameter group cannot be accurately determined. Therefore, the present application compares the blade performance prediction value with the initial performance benchmark by configuring the use of a multi-objective optimization algorithm, and based on the comparison result, multiple candidate design parameter groups are selected from each design parameter group, thereby improving the accuracy of the determined blade optimization parameter group.

[0091] S105, based on each candidate design parameter group, perform computational fluid dynamics simulation to obtain simulation performance parameters of each candidate design parameter group, and determine the candidate design parameter group corresponding to the optimal simulation performance parameter as the blade optimization parameter group.

[0092] It should be noted that the present application performs computational fluid dynamics simulation based on each candidate design parameter group, thereby further verifying the performance of each candidate design parameter group, thereby ensuring that the finally determined blade optimization parameter group is a group of optimization parameters that optimizes the performance of the to-be-optimized axial flow compressor blade, thereby improving the prediction accuracy of the small sample training proxy model for blade optimization parameters compared to the prior art.

[0093] This application achieves the determination of the optimization object and reference standard by configuring and obtaining candidate optimization design parameters affecting the aerodynamic performance of the axial compressor blade to be optimized, and initial performance benchmarks obtained from computational fluid dynamics simulation based on the original blade profile. Subsequently, by configuring and utilizing a data sampling algorithm based on the design parameters to be optimized and their adjustable range, multiple design parameter groups are generated. This reduces the risk of invalid parameters in the design parameter groups by utilizing the adjustable range, thereby improving the accuracy of subsequent prediction and optimization. Furthermore, by configuring the training data of the master-slave combined surrogate model to include at least the flow field characteristic parameters obtained from computational fluid dynamics simulation based on the original blade profile, the master-slave combined surrogate model can predict the performance of the design parameter groups based on the flow field characteristics captured during training. Subsequently, by configuring and utilizing a multi-objective optimization algorithm, the predicted performance values ​​of each blade profile are compared with the initial performance benchmark, and multiple candidate design parameter groups are selected from each design parameter group based on the comparison results. This reduces the computational load of subsequent computational fluid dynamics simulations and improves the simulation accuracy of the subsequently obtained simulation performance parameters. Finally, by configuring computational fluid dynamics simulation based on each alternative design parameter group, the simulation performance parameters of each alternative design parameter group are obtained, and the alternative design parameter group corresponding to the optimal simulation performance parameters is determined as the airfoil optimization parameter group, thereby improving the prediction accuracy of the finally determined airfoil optimization parameter group.

[0094] In one possible implementation, the training process of the aforementioned master-slave combined agent model includes:

[0095] Obtain the sweep parameters of multiple control points of the original airfoil. The sweep parameters of a set of control points include the sweep values ​​of each control point of the Bezier curve of the axial compressor blade to be optimized of the original airfoil.

[0096] Computational fluid dynamics simulations were performed based on the sweep parameters of each set of control points to obtain the stable operating range parameters and fluid characteristic parameters corresponding to the sweep parameters of each set of control points.

[0097] The dataset including the sweep parameters of each group of control points with added labels is determined as the first training set, and the labels are the stable working range parameters corresponding to the sweep parameters of the control points; the dataset including each fluid characteristic parameter is determined as the second training set.

[0098] The initial primary and secondary agent models are trained based on the first and second training sets to obtain the primary and secondary combined agent model. The input of the primary and secondary combined agent model is multiple design parameter sets generated based on the design parameters to be optimized and the adjustable range of the parameters of the axial compressor blade to be optimized. The output is the blade performance prediction value corresponding to each design parameter set.

[0099] It should be noted that, in practical application scenarios, the aforementioned stable operating range parameter M... SWThe near-stall point mass flow M NC and the near-stall point mass flow M NS , as shown in the formula: M SW =M NC -M NS .

[0100] In one possible implementation, the initial master-slave agent model is trained based on the first training set and the second training set to obtain a master-slave combined agent model, including,

[0101] initializing the initial master-slave agent model, the master-slave agent model including a master model, a slave model, a fusion layer, and a linear transformation layer;

[0102] inputting the first training set into the master model to obtain initial blade profile performance prediction results of the sweep parameters of each group of control points output by the master model;

[0103] inputting the second training set into the slave model to obtain auxiliary blade profile performance prediction results of the sweep parameters of each group of control points output by the slave model;

[0104] inputting each initial blade profile performance prediction result and each auxiliary blade profile performance prediction result into the fusion layer, so that the fusion layer sums the initial blade profile performance prediction result and the auxiliary blade profile performance prediction result corresponding to the sweep parameters of each group of control points, and calls the linear transformation layer to adjust the sum weight, to obtain blade profile performance prediction values of the sweep parameters of each group of control points, and based on the blade profile performance prediction values and the initial performance benchmark, to adjust the parameters of the master model and the slave model to obtain a master-slave combined agent model.

[0105] It should be noted that in actual application scenarios, the above-mentioned initial master-slave agent model can be a model constructed based on a Multilayer Perceptron (MLP) architecture. Since the number of training data in the first training set is limited in a small sample scenario, and fewer training data can cause ordinary MLP architecture to have overfitting risk. Therefore, the above-mentioned master model can be configured as a regularized MLP architecture, so as to improve the data generalization ability of the master model in a small sample scenario by using the regularization ability, avoid the risk of missing capturing features due to overfitting of the master model, and improve the output precision of the master model. And because the input dimension of the slave model is small, in order to improve the overall prediction efficiency of the master-slave combined agent model and avoid the risk of prediction accuracy decline caused by the increase of fitting degree due to the complexity of the slave model structure, the above-mentioned slave model can be configured as a lightweight MLP architecture with a lightweight feature enhancement channel, so as to ensure the output accuracy of the slave model while avoiding the risk of increasing the overall fitting degree of the master-slave combined agent model.

[0106] In one possible implementation, the mathematical representation of the above-mentioned linear transformation layer can be as shown in the following formula: ,in, It is the summation weight of the initial or auxiliary airfoil performance prediction results. It is a weight matrix. It is a bias vector. Each (The sum of the initial and auxiliary airfoil performance predictions corresponding to the sweep parameters of a set of control points). During the process of adjusting the summation weights by calling the linear transformation layer, the above is first solved based on the initial and auxiliary airfoil performance predictions from the input fusion layer. And call the above linear transformation layer to solve Then the Determined as above During the solution process The summation weights of main, and... The summation weights of aux are set to Then, the sum of the initial and auxiliary airfoil performance prediction results corresponding to the sweep parameters of each control point is calculated. Since the importance of the output results of the main model and the sub-model differs under different input conditions, this application improves the output accuracy of the blade performance prediction value by configuring the above-mentioned linear transformation layer to dynamically adjust the summation weights.

[0107] In one possible implementation, the architecture diagram of the initial master-slave agent model described above is as follows: Figure 3 As shown, the main model inputs the initial airfoil performance prediction results into the fusion layer, and the sub-model inputs the auxiliary airfoil performance prediction results into the fusion layer. The fusion layer sums the initial airfoil performance prediction results and the auxiliary airfoil performance prediction results corresponding to the sweep parameters of a set of control points, and calls the linear transformation layer to adjust the summation weights during the summation process, and outputs the airfoil performance prediction values ​​of the sweep parameters of each set of control points.

[0108] In one possible implementation, the predicted airfoil performance values ​​include at least the predicted stable operating range and the predicted peak efficiency condition, and the initial performance baseline includes at least the initial stable operating range and the initial peak efficiency condition. A multi-objective optimization algorithm is used to compare the predicted airfoil performance values ​​with the initial performance baseline, and based on the comparison results, multiple candidate design parameter sets are selected from each set of design parameters, including:

[0109] Using a multi-objective optimization algorithm, the predicted range of each stable operating condition is compared with the initial stable operating range, and the predicted peak efficiency condition is compared with the initial peak efficiency condition. The design parameter group corresponding to the predicted airfoil performance values ​​that are not less than the initial stable operating range and whose predicted peak efficiency condition is not less than the initial peak efficiency condition is determined as the alternative design parameter group.

[0110] It should be noted that in the actual application scenario, the peak efficiency in the peak efficiency prediction working condition and the initial peak efficiency working condition can be calculated by the following formula: wherein m is the mass flow rate; η PE is the peak efficiency value under the corresponding working condition; π t is the peak efficiency point pressure ratio; T t，in is the total temperature at the inlet; P in is the total pressure at the inlet; c p is the static pressure coefficient; and k is the specific heat ratio.

[0111] In one possible implementation, the type of fluid characteristic parameter is at least one or more of a high-entropy region value, an axial momentum of the highest efficiency point to be optimized axial flow compressor blade, and a position of an aerodynamic load center.

[0112] It should be noted that in the actual application scenario, the high-entropy region value can be based on an average value of a high-entropy region tail width at different blade heights (such as 25%, 50%, and 75% blade heights) in an entropy diagram of a flow field based on simulation results, wherein a schematic diagram of a blade high-entropy region can be as shown in Figure 4 , the high-entropy region is a region that characterizes the degree of irreversible energy dissipation and flow disorder of the airflow after flowing along the surface of the blade from the leading edge of the blade to the trailing edge of the blade, Figure 4 the medium-high-entropy region value is a default Figure 4 medium-high-entropy region is a high-entropy region with an entropy value greater than a preset threshold, and a width value corresponding to an average value of a high-entropy region tail width of the high-entropy region. The axial momentum of the highest efficiency point to be optimized axial flow compressor blade can be obtained by calculation based on the following formula, based on the axial plane corresponding to the main flow axial momentum in the simulation process, the axial velocity at the radial range of 20% to 100% blade height, the fluid density, and the reference flow rate, and two grid spacings in the circumferential direction: wherein M Z,in is the axial momentum of the main flow; V z is the axial velocity; p is the fluid density; and m is the reference flow rate. The position of the aerodynamic load center is obtained by the following method: starting from the leading edge of the blade profile, the pressure difference between the suction surface and the pressure surface is integrated, and when the integral value reaches half of the total amount of the pressure difference of the entire section, the corresponding axial coordinate is the position of the aerodynamic load center. That is, at a certain position of the blade profile, the pressure difference integral from the leading edge of the blade profile to the position accounts for 50% of the total amount of the pressure difference of the entire blade profile section.

[0113] In one possible implementation, the method further comprises:

[0114] In the absence of an alternative design parameter set, the design parameter set corresponding to each blade profile performance prediction value is used as supplementary training data to optimize and train the main-vice combined surrogate model;

[0115] In the case where the main-vice combined surrogate model is optimized and trained, the operation step of sequentially inputting each design parameter set into the main-vice combined surrogate model to obtain the blade profile performance prediction value corresponding to each design parameter set is performed.

[0116] It should be noted that in actual application scenarios, the present application optimizes and trains the main-vice combined surrogate model using supplementary training data in the absence of an alternative design parameter set, thereby dynamically optimizing the main-vice combined surrogate model in a small sample scenario and improving the output accuracy of the main-vice combined surrogate model.

[0117] The present application has verified the optimization effect of the axial flow compressor blade profile optimization method based on the main-vice surrogate model provided by the present application through numerical simulation of a transonic axial flow compressor with 16 channels. To save computational cost, a single channel is selected for numerical simulation. AutoGrid5 in NUMECA software is used to divide the blade channel grid. The divided compressor grid is imported into the CFX preprocessor, the inlet conditions are total temperature and total pressure, the outlet boundary condition is static pressure outlet, and the turbulence model is SST (Shear Stress Transfer). Then, CFD-post is used for post-processing of the numerical calculation results to obtain the optimization results. Through comparative analysis of the experimental results, the optimization results (optimized blade profile) and the flow ratio-pressure ratio characteristic curve of the original blade profile (prototype) are obtained as shown in Figure 5 The optimization results (optimized blade profile) and the flow ratio-isentropic efficiency characteristic curve of the original blade profile (prototype) are shown in Figure 6 The margin increase value in the above two figures represents the stall margin improvement of the optimization result compared with the original blade profile. The two broken lines in the two figures are the overall stable working ranges of the original blade profile and the optimized blade profile, respectively. The larger the range is, the better it is. However, while the x-coordinate (flow coefficient) range is expanded, the y-coordinate total pressure ratio and isentropic efficiency cannot be greatly reduced. The research results show that: compared with the traditional surrogate model, the main-vice combined surrogate model based on the small sample library in the present application improves the prediction accuracy by 61.2%. The optimized blade profile has a forward sweep of 2.91° at 95% blade height and a backward sweep of 2.63° at 50% blade height. The optimized blade profile has a peak efficiency improvement of 0.5% and a stable working range increase of 17.7%. The applicability of the main-vice combined surrogate model in the performance optimization design of the compressor is verified. The second aspect of the present application provides an axial flow compressor blade profile optimization system as shown in Figure 7 The axial flow compressor blade profile optimization system comprises:

[0118] The parameter obtaining module 701 is configured to obtain an initial performance benchmark and an optimization design parameter of an optimization axial compressor blade, the optimization design parameter being an adjustable geometry parameter of the blade affecting aerodynamic performance of the blade, and the initial performance benchmark being a result parameter obtained by performing computational fluid dynamics simulation based on an original blade profile of the optimization axial compressor blade.

[0119] The parameter generating module 702 is configured to generate a plurality of design parameter groups based on the optimization design parameter and a parameter adjustable range by using a data sampling algorithm.

[0120] The performance prediction module 703 is configured to input each design parameter group into the main-sub combined proxy model in sequence to obtain a blade profile performance prediction value corresponding to each design parameter group, and the training data of the main-sub combined proxy model at least includes a flow field characteristic parameter obtained by performing computational fluid dynamics simulation based on the original blade profile.

[0121] The performance comparison module 704 is configured to compare each blade profile performance prediction value with the initial performance benchmark by using a multi-objective optimization algorithm, and select a plurality of candidate design parameter groups from the design parameter groups based on a comparison result.

[0122] The parameter output module 705 is configured to perform computational fluid dynamics simulation based on each candidate design parameter group to obtain a simulation performance parameter of each candidate design parameter group, and determine a candidate design parameter group corresponding to an optimal simulation performance parameter as a blade profile optimization parameter group.

[0123] In a possible implementation, the optimization system of the axial compressor blade provided in the second aspect of the present application further includes a model training module, which is configured to:

[0124] obtain a plurality of sets of sweep parameters of control points of the original blade profile, a sweep parameter of a set of control points including sweep values of control points of a Bezier curve of the optimization axial compressor blade of the original blade profile;

[0125] perform computational fluid dynamics simulation based on the sweep parameters of each set of control points to obtain a stable working range parameter and a fluid characteristic parameter corresponding to each sweep parameter of each set of control points;

[0126] determine a data set including the sweep parameters of each set of control points added with labels as a first training set, and a content of the label being the stable working range parameter corresponding to the sweep parameter of the control point; and determine a data set including the fluid characteristic parameters as a second training set;

[0127] The initial master-slave agent model is trained based on the first training set and the second training set to obtain a master-slave combined agent model, and an input of the master-slave combined agent model is a plurality of design parameter groups generated based on the to-be-optimized design parameters of the to-be-optimized axial flow compressor blade and a parameter adjustable range, and an output is a blade profile performance prediction value corresponding to each design parameter group.

[0128] In a possible implementation, the model training module is configured to, in the process of training the initial master-slave agent model based on the first training set and the second training set to obtain the master-slave combined agent model:

[0129] The initial master-slave agent model is initialized, and the master-slave agent model includes a master model, a slave model, a fusion layer, and a linear transformation layer.

[0130] The first training set is input into the master model to obtain initial blade profile performance prediction results of the sweep parameters of each group of control points output by the master model.

[0131] The second training set is input into the slave model to obtain auxiliary blade profile performance prediction results of the sweep parameters of each group of control points output by the slave model.

[0132] The initial blade profile performance prediction results and the auxiliary blade profile performance prediction results are input into the fusion layer, so that the fusion layer sums the initial blade profile performance prediction results and the auxiliary blade profile performance prediction results corresponding to the sweep parameters of each group of control points, and calls the linear transformation layer to adjust the sum weight to obtain blade profile performance prediction values of the sweep parameters of each group of control points, and based on the blade profile performance prediction values and the initial performance benchmark, the master model and the slave model are adjusted in parameters to obtain the master-slave combined agent model.

[0133] In a possible implementation, the performance comparison module 704 is configured to:

[0134] Each stable working prediction range is compared with the initial stable working range by using a multi-objective optimization algorithm, each peak efficiency prediction working condition is compared with the initial peak efficiency working condition, and a design parameter group corresponding to a blade profile performance prediction value, in which the stable working prediction range is not less than the initial stable working range and the peak efficiency prediction working condition is not less than the initial peak efficiency working condition, is determined as a candidate design parameter group, the blade profile performance prediction value at least includes the stable working prediction range and the peak efficiency prediction working condition, and the initial performance benchmark at least includes the initial stable working range and the initial peak efficiency working condition.

[0135] In a possible implementation, the type of the fluid characteristic parameter is at least one or more of a high-entropy region value, an axial momentum of the to-be-optimized axial flow compressor blade at a highest efficiency point, and a position of an aerodynamic load center.

[0136] In a possible implementation, the performance comparison module 704 is further configured to:

[0137] In the absence of an alternative design parameter set, the design parameter set corresponding to each blade profile performance prediction value is used as supplementary training data to optimize and train the main-vice combined surrogate model using the supplementary training data;

[0138] In the case where the main-vice combined surrogate model is optimized and trained, the performance prediction module is triggered to perform the operation step of sequentially inputting each design parameter set into the main-vice combined surrogate model to obtain the blade profile performance prediction value corresponding to each design parameter set.

[0139] The third aspect of the present application provides a computer program product, which comprises computer readable instructions, when the computer readable instructions are run on an electronic device, the electronic device implements the axial flow compressor blade profile optimization method based on the main-vice surrogate model of the first aspect or any implementation manner of the first aspect.

[0140] The fourth aspect of the present application provides an electronic device, which comprises at least one processor and a memory connected with the processor, wherein:

[0141] The memory is used to store a computer program;

[0142] The processor is used to execute the computer program, so that the electronic device can implement the axial flow compressor blade profile optimization method based on the main-vice surrogate model of the first aspect or any implementation manner of the first aspect.

[0143] The fifth aspect of the present application provides a computer storage medium, which carries one or more computer programs, when the one or more computer programs are executed by an electronic device, the electronic device can implement the axial flow compressor blade profile optimization method based on the main-vice surrogate model of the first aspect or any implementation manner of the first aspect.

[0144] The structure schematic diagram of the electronic device provided by the fourth aspect of the present application is shown in Figure 8 The electronic device in the embodiments of the present application can include but is not limited to fixed terminals such as mobile phones, notebook computers, PDAs (personal digital assistants), PADs (tablets), desktop computers and the like. Figure 8 The electronic device shown is only an example, and should not bring any limitation to the functions and use range of the embodiments of the present application.

[0145] As Figure 8As shown, the electronic device can include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 801 that can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 802 or loaded into a random access memory (RAM) 803 from a storage device 808. In a state in which the electronic device is powered on, various programs and data required for operation of the electronic device are also stored in the RAM 803. The processing device 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0146] In general, the following devices can be connected to the I / O interface 805: input devices 806 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 807 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 808 including, for example, a memory card, a hard disk, etc.; and communication devices 809. The communication devices 809 can allow the electronic device to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 8 An electronic device having various devices is shown, but it is understood that all of the shown devices are not required to be implemented or present. More or less devices can alternatively be implemented or present.

[0147] It should be further noted that the above-described device embodiments are merely illustrative, wherein the units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on a plurality of network units. Part or all of the modules can be selected according to actual needs to achieve the purpose of the present embodiment. In addition, the connection relationship between the modules in the device embodiment provided in the present application indicates that there is a communication connection between them, which can be implemented as one or more communication buses or signal lines.

[0148] Those skilled in the art can clearly understand that the application can be implemented by means of software plus necessary universal hardware, and of course can also be implemented by means of dedicated hardware including special-purpose integrated circuits, special-purpose CPUs, special-purpose memories, special-purpose components, etc. Generally, any function completed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure for implementing the same function can also be various, such as analog circuits, digital circuits, or special-purpose circuits, etc. However, for the present application, software program implementation is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which is stored in a readable storage medium, such as a floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a magnetic disk or an optical disk, etc., and includes a plurality of instructions for causing a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in the various embodiments of the present application.

[0149] In the above embodiments, all or part of the embodiments can be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments can be implemented in the form of a computer program product.

[0150] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of the present application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, training device or data center to another website, computer, training device or data center through wired (such as coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer-readable storage medium can be any available medium that can be stored by a computer or a data storage device such as a training device, a data center, etc. integrated with one or more available media. The available media can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium (such as a solid state disk (SSD)), etc.

Claims

1. A method for axial flow compressor profile optimization based on master-slave agent model, characterized in that, The application relates to an optimization method for an axial-flow compressor blade, and belongs to the field of optimization design of axial-flow compressor blades. The method comprises the following steps: obtaining design parameters to be optimized of an axial-flow compressor blade to be optimized and initial performance benchmarks, wherein the design parameters to be optimized are adjustable geometric parameters of the blade which affect the aerodynamic performance of the blade, and the initial performance benchmarks are result parameters obtained through computational fluid dynamics simulation based on an original blade profile of the axial-flow compressor blade to be optimized; generating a plurality of design parameter groups based on the design parameters to be optimized and adjustable parameter ranges through a data sampling algorithm; inputting each of the design parameter groups into a main-vice combined proxy model in sequence to obtain blade profile performance prediction values corresponding to each of the design parameter groups, wherein training data of the main-vice combined proxy model at least comprises flow field characteristic parameters obtained through computational fluid dynamics simulation based on the original blade profile; comparing each of the blade profile performance prediction values with the initial performance benchmarks through a multi-objective optimization algorithm, and screening a plurality of alternative design parameter groups from the design parameter groups based on comparison results; performing computational fluid dynamics simulation based on each of the alternative design parameter groups to obtain simulation performance parameters of each of the alternative design parameter groups, and determining an alternative design parameter group corresponding to optimal simulation performance parameters as a blade profile optimization parameter group; the training process of the main-vice combined proxy model comprises the following steps: obtaining a plurality of sets of control point sweep parameters of the original blade profile, wherein a set of control point sweep parameters comprises sweep values of control points of a Bezier curve of the axial-flow compressor blade to be optimized of the original blade profile; performing computational fluid dynamics simulation based on each set of control point sweep parameters to obtain stable working range parameters and flow field characteristic parameters corresponding to each set of control point sweep parameters; determining a data set comprising each set of control point sweep parameters added with labels as a first training set, wherein the content of the labels is the stable working range parameters corresponding to the control point sweep parameters; and determining a data set comprising each of the flow field characteristic parameters as a second training set; 2. The axial flow compressor profile optimization method based on the primary-secondary agent model according to claim 1, characterized in that, training an initial main-vice proxy model based on the first training set and the second training set to obtain the main-vice combined proxy model, wherein the input of the main-vice combined proxy model is a plurality of design parameter groups generated based on the design parameters to be optimized of the axial-flow compressor blade to be optimized and adjustable parameter ranges, and the output is blade profile performance prediction values corresponding to each of the design parameter groups. The training of the initial main-vice proxy model based on the first training set and the second training set to obtain the main-vice combined proxy model comprises the following steps: initializing the initial main-vice proxy model, wherein the main-vice proxy model comprises a main model, a vice model, a fusion layer and a linear transformation layer; inputting the first training set into the main model to obtain initial blade profile performance prediction results of each set of control point sweep parameters output by the main model; inputting the second training set into the vice model to obtain auxiliary blade profile performance prediction results of each set of control point sweep parameters output by the vice model; and The initial blade profile performance prediction results and the auxiliary blade profile performance prediction results are input into the fusion layer, so that the fusion layer sums the initial blade profile performance prediction results and the auxiliary blade profile performance prediction results corresponding to the sweep parameters of each group of control points, adjusts the sum weight by calling the linear transformation layer, obtains the blade profile performance prediction values of the sweep parameters of each group of control points, and adjusts the parameters of the primary model and the secondary model based on the blade profile performance prediction values and the initial performance benchmark, to obtain the primary-secondary combined proxy model.

3. The axial flow compressor profile optimization method based on the primary-secondary agent model according to claim 1, characterized in that, The blade profile performance prediction values at least include stable working prediction ranges and peak efficiency prediction working conditions, the initial performance benchmark at least includes initial stable working ranges and initial peak efficiency working conditions, and the multi-objective optimization algorithm is used to compare each blade profile performance prediction value with the initial performance benchmark, and based on the comparison result, a plurality of alternative design parameter groups are screened from each design parameter group, including: The multi-objective optimization algorithm is used to compare each stable working prediction range with the initial stable working range, compare each peak efficiency prediction working condition with the initial peak efficiency working condition, and the design parameter group corresponding to the blade profile performance prediction value that is not less than the initial stable working range and not less than the initial peak efficiency working condition is determined as the alternative design parameter group.

4. The axial flow compressor profile optimization method based on the primary-secondary agent model according to claim 1, characterized in that, The type of the flow field characteristic parameter is at least one or more of a high-entropy region value, a highest efficiency point, an axial momentum of the blade of the to-be-optimized axial flow compressor, and a position of a center of aerodynamic load.

5. The axial flow compressor profile optimization method based on the primary-secondary agent model according to claim 3, characterized in that, Further comprising: In the absence of the alternative design parameter group, the design parameter group corresponding to each blade profile performance prediction value is used as supplementary training data, and the primary-secondary combined proxy model is optimized and trained by using the supplementary training data; In the case where the primary-secondary combined proxy model is optimized and trained, the operation step of sequentially inputting each design parameter group into the primary-secondary combined proxy model to obtain the blade profile performance prediction value corresponding to each design parameter group is performed.

6. A system for optimization of axial flow compressor airfoil based on primary-secondary agent model, characterized in that, Comprising: A parameter obtaining module is configured to obtain to-be-optimized design parameters of a to-be-optimized axial flow compressor blade and an initial performance benchmark, the to-be-optimized design parameters being blade adjustable geometric parameters affecting blade aerodynamic performance, and the initial performance benchmark being a result parameter obtained by computational fluid dynamics simulation based on an original blade profile of the to-be-optimized axial flow compressor blade; A parameter generating module is configured to generate a plurality of design parameter groups based on the to-be-optimized design parameters and a parameter adjustable range by using a data sampling algorithm; A performance prediction module is configured to sequentially input each design parameter group into a primary-secondary combined proxy model to obtain a blade profile performance prediction value corresponding to each design parameter group, and training data of the primary-secondary combined proxy model at least including flow field characteristic parameters obtained by computational fluid dynamics simulation based on an original blade profile. a performance comparison module configured to compare each of the blade profile performance prediction values with the initial performance benchmark by using a multi-objective optimization algorithm, and to select a plurality of candidate design parameter groups from the design parameter groups based on comparison results; a parameter output module configured to perform computational fluid dynamics simulation based on each of the candidate design parameter groups, to obtain simulation performance parameters of each of the candidate design parameter groups, and to determine the candidate design parameter group corresponding to the optimal simulation performance parameter as the blade profile optimization parameter group; the training process of the main-vice combined surrogate model comprises: obtaining a plurality of groups of control point sweep parameters of the original blade profile, wherein the sweep parameter of each group of control points comprises a sweep value of each control point of a Bezier curve of the axial compressor blade to be optimized of the original blade profile; performing computational fluid dynamics simulation based on the sweep parameter of each group of control points to obtain a stable working range parameter and a flow field characteristic parameter corresponding to each sweep parameter of each group of control points; determining a data set comprising the sweep parameter of each group of control points added with a label as a first training set, wherein the content of the label is the stable working range parameter corresponding to the sweep parameter of the control point; and determining a data set comprising each of the flow field characteristic parameters as a second training set; training an initial main-vice surrogate model based on the first training set and the second training set to obtain the main-vice combined surrogate model, wherein the input of the main-vice combined surrogate model is a plurality of design parameter groups generated based on the design parameter to be optimized of the axial compressor blade to be optimized and the parameter adjustable range, and the output is a blade profile performance prediction value corresponding to each design parameter group.

7. A computer program product, characterised in that, A computer readable instruction is included, and when the computer readable instruction is run on an electronic device, the electronic device is enabled to implement the axial compressor blade profile optimization method based on the main-vice surrogate model according to any one of claims 1 to 5.

8. An electronic device, comprising: An electronic device comprises at least one processor and a memory connected with the processor, wherein: the memory is configured to store a computer program; the processor is configured to execute the computer program to enable the electronic device to implement the axial compressor blade profile optimization method based on the main-vice surrogate model according to any one of claims 1 to 5.

9. A computer storage medium, characterized in that The storage medium carries one or more computer programs, and when the one or more computer programs are executed by an electronic device, the electronic device is enabled to implement the axial compressor blade profile optimization method based on the main-vice surrogate model according to any one of claims 1 to 5.

Images