Fourier neural operator-based flow field prediction model training method and device, computer device, and storage medium

Abstract

The application relates to a flow field prediction model training method and device based on a Fourier neural operator, computer equipment and a storage medium. The flow field prediction model is constructed based on a Fourier neural operator and comprises a Fourier layer and an output layer. The method comprises the following steps: determining an original flow field prediction model corresponding to a target working condition, wherein the original flow field prediction model is obtained by training based on working condition conditions of an original working condition; generating simulation flow field data corresponding to a target blade row under the target working condition; obtaining predicted flow field data by using the original flow field prediction model; determining a loss value according to the difference between the predicted flow field data and the simulation flow field data; adjusting the output layer of the original flow field prediction model based on the loss value to obtain a trained flow field prediction model corresponding to the target working condition. The method can improve the training efficiency of the flow field prediction model.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aero-engines, and particularly relates to a flow field prediction model training method and device based on a Fourier neural operator, a computer device and a storage medium. BACKGROUND

[0002] Compressor blades are a key component in modern aero-engine design, and they significantly affect the aerodynamic performance of the engine, including compression efficiency, airflow loss and stall behavior. In order to optimize the aerodynamic characteristics of the compressor blades, the designed blades usually need to be simulated multiple times in the engineering design process.

[0003] In related technologies, a Reynolds-Averaged Navier-Stokes (RANS) equation solver is a mainstream tool for simulating complex airflow behavior. However, the application of the RANS model requires a large amount of computing resources and time, especially when dealing with transonic flow fields such as supercritical airfoils, the simulation cost is extremely high, which seriously affects the design cycle. Moreover, the RANS model does not have the generalization ability for different working conditions, and usually needs to be simulated from scratch when simulating the flow field under different working conditions, resulting in low efficiency of flow field simulation.

[0004] Therefore, it is necessary to provide a flow field prediction method capable of improving the efficiency of flow field simulation to quickly obtain the total pressure loss of the compressor flow field. SUMMARY

[0005] Therefore, it is necessary to provide a flow field prediction method capable of improving the efficiency of flow field simulation to quickly obtain the total pressure loss of the compressor flow field.

[0006] In a first aspect, the present application provides a flow field prediction model training method based on a Fourier neural operator. The flow field prediction model is constructed based on a Fourier neural operator and includes a Fourier layer and an output layer. The method comprises:

[0007] determining an original flow field prediction model corresponding to a target working condition, wherein the original flow field prediction model is trained based on working condition conditions of an original working condition;

[0008] generating simulation flow field data corresponding to a target cascade under the target working condition according to working condition conditions of the target working condition;

[0009] performing feature extraction processing on the working condition conditions through the Fourier layer of the original flow field prediction model to obtain flow field feature data, and performing flow field reconstruction processing on the flow field feature data through the output layer of the original flow field prediction model to obtain predicted flow field data;

[0010] A loss value is determined according to a difference between the predicted flow field data and the simulated flow field data, an output layer of the original flow field prediction model is adjusted based on the loss value, and a trained flow field prediction model corresponding to the target working condition is obtained.

[0011] In one of the embodiments, the original flow field prediction model is trained based on an initial flow field prediction model, and the initial flow field prediction model adjusts the Fourier layer and the output layer based on a loss value during the training process.

[0012] In one of the embodiments, the feature extraction processing of the working condition by the Fourier layer of the original flow field prediction model obtains flow field feature data, including:

[0013] The working condition is converted to a frequency domain, and a preset number of low-frequency components with a frequency less than a preset threshold are extracted from the working condition converted to the frequency domain;

[0014] The low-frequency components are subjected to frequency domain feature modulation processing;

[0015] The low-frequency components subjected to the frequency domain feature modulation are converted to a spatial domain to obtain flow field feature data.

[0016] In one of the embodiments, the determination of the loss value according to the difference between the predicted flow field data and the simulated flow field data includes:

[0017] A physical residual loss of the predicted flow field data is determined based on the Navier-Stokes equation;

[0018] A flow field prediction loss is determined based on the difference between the predicted flow field data and the simulated flow field data;

[0019] A loss value is determined based on the physical residual loss and the flow field prediction loss.

[0020] In one of the embodiments, the method further includes:

[0021] The flow field feature data is subjected to total pressure loss prediction processing by the output layer of the original flow field prediction model to obtain predicted total pressure loss data;

[0022] The determination of the loss value based on the physical residual loss and the flow field prediction loss includes:

[0023] A total pressure loss prediction loss is determined according to a difference between the predicted total pressure loss data and simulated total pressure loss data;

[0024] A loss value is determined based on the physical residual loss, the flow field prediction loss, and the total pressure loss prediction loss.

[0025] In a second aspect, the application also provides a flow field prediction method based on a Fourier neural operator, the method comprising:

[0026] determining a working condition of a target working condition;

[0027] inputting the working condition into a flow field prediction model to obtain predicted flow field data corresponding to a target cascade under the target working condition;

[0028] The flow field prediction model is trained by the method of any of the preceding embodiments.

[0029] In a third aspect, the application also provides a flow field prediction model training device based on a Fourier neural operator. The flow field prediction model is constructed based on a Fourier neural operator and includes a Fourier layer and an output layer. The device comprises:

[0030] A determination module is configured to determine an original flow field prediction model corresponding to a target working condition, which is trained based on a working condition of an original working condition;

[0031] A generation module is configured to generate simulation flow field data corresponding to a target cascade under the target working condition according to the working condition of the target working condition;

[0032] A reconstruction module is configured to perform feature extraction processing on the working condition through the Fourier layer of the original flow field prediction model to obtain flow field feature data, and perform flow field reconstruction processing on the flow field feature data through the output layer of the original flow field prediction model to obtain predicted flow field data;

[0033] A training module is configured to determine a loss value based on the difference between the predicted flow field data and the simulation flow field data, adjust the output layer of the original flow field prediction model based on the loss value, and obtain a trained flow field prediction model corresponding to the target working condition.

[0034] In one embodiment, the original flow field prediction model is trained based on an initial flow field prediction model, and the initial flow field prediction model adjusts the Fourier layer and the output layer based on the loss value during the training process.

[0035] In one embodiment, the reconstruction module is further configured to:

[0036] convert the working condition to a frequency domain, and extract a preset number of low-frequency components with a frequency less than a preset threshold from the converted working condition in the frequency domain;

[0037] perform frequency domain feature modulation processing on each of the low-frequency components;

[0038] Convert each low-frequency component after frequency domain feature modulation to the spatial domain to obtain flow field feature data.

[0039] In one of the embodiments, the training module is further configured to:

[0040] determine a physical residual loss of the predicted flow field data based on the Navier-Stokes equation;

[0041] determine a flow field prediction loss based on a difference between the predicted flow field data and the simulated flow field data;

[0042] determine a loss value based on the physical residual loss and the flow field prediction loss.

[0043] In one of the embodiments, the reconstruction module is further configured to:

[0044] perform total pressure loss prediction processing on the flow field feature data through an output layer of the original flow field prediction model to obtain predicted total pressure loss data;

[0045] The training module is further configured to:

[0046] determine a total pressure loss prediction loss based on a difference between the predicted total pressure loss data and simulated total pressure loss data;

[0047] determine a loss value based on the physical residual loss, the flow field prediction loss, and the total pressure loss prediction loss.

[0048] In a fourth aspect, the present application further provides a flow field prediction device based on a Fourier neural operator, the device comprising:

[0049] a determination module configured to determine a working condition of a target working condition;

[0050] an input module configured to input the working condition to a flow field prediction model to obtain predicted flow field data corresponding to a target blade row under the target working condition;

[0051] The flow field prediction model is trained by the method of any of the preceding embodiments.

[0052] In a fifth aspect, the present application further provides a computer device. The computer device comprises a memory and a processor, the memory stores a computer program, and the processor implements any of the preceding methods when executing the computer program.

[0053] In a sixth aspect, the present application further provides a computer readable storage medium. The computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement any of the preceding methods.

[0054] In a seventh aspect, the present application also provides a computer program product. The computer program product comprises a computer program which, when executed by a processor, implements any of the above methods.

[0055] The flow field prediction model training method, device, computer device and storage medium based on the Fourier neural operator described above apply the Fourier neural operator to flow field prediction, divide the flow field prediction model into a Fourier layer adopting a Fourier neural operator architecture and an output layer, and obtain a flow field training model for a target working condition by fine-tuning the output layer of an original flow field prediction model trained under other working conditions using working condition conditions under the target working condition. Since it is easier to capture global features and long-term dependencies of a flow field in the frequency domain, and the global features of a flow field under different working conditions have certain similarity, the flow field prediction model is constructed using the Fourier neural operator, the parameters of the Fourier layer do not need to be adjusted when the model is migrated to other working conditions, the parameters of the output layer can be adjusted to obtain a flow field reconstruction model conforming to the characteristics of the target working condition, the speed of fine-tuning and training the model under other working conditions can be accelerated, and the flow field prediction model training efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 FIG. 1 is a flowchart of a flow field prediction model training method based on a Fourier neural operator in an embodiment;

[0057] Figure 2 FIG. 2 is a flowchart of a flow field feature data extraction method in an embodiment;

[0058] Figure 3 FIG. 3 is a flowchart of step 108 in an embodiment;

[0059] Figure 4 FIG. 4 is a flowchart of a flow field prediction method based on a Fourier neural operator in an embodiment;

[0060] Figure 5 FIG. 5 is a structural block diagram of a flow field prediction model training device based on a Fourier neural operator in an embodiment;

[0061] Figure 6 FIG. 6 is a structural block diagram of a flow field prediction device based on a Fourier neural operator in an embodiment;

[0062] Figure 7 FIG. 7 is an internal structure diagram of a computer device in an embodiment. DETAILED DESCRIPTION

[0063] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not intended to limit the present application.

[0064] In one embodiment, as shown in Figure 1 A flow field prediction model training method based on a Fourier neural operator is provided. This embodiment illustrates the method applied to a server. It should be understood that the method can also be applied to a terminal, and can also be applied to a system including a terminal and a server, and can be realized through the interaction of the terminal and the server. In this embodiment, the method includes the following steps:

[0065] Step 102, determining an original flow field prediction model corresponding to the target working condition, the original flow field prediction model being trained based on the working condition of the original working condition.

[0066] In the embodiments of the present application, one or more representative working conditions can be used as original working conditions, and the original flow field prediction models corresponding to these original working conditions can be trained based on the working conditions of the original working conditions. When it is necessary to train a flow field prediction model for other working conditions, the flow field prediction model corresponding to the other working conditions can be quickly trained by fine-tuning the original flow field prediction model.

[0067] The original flow field prediction model corresponding to the original working condition similar to the target working condition can be selected as the original flow field prediction model corresponding to the target working condition. Similar can refer to similar boundary conditions, similar fluid physical properties, similar forms of the working condition, and the like, which are not limited in the embodiments of the present application.

[0068] Step 104, generating simulation flow field data corresponding to the target cascade under the target working condition according to the working condition of the target working condition.

[0069] In the embodiments of the present application, the working condition refers to the relevant parameters that determine the steady flow field in the target cascade, which can include cascade parameters such as blade angle, chord length, and pitch of the target cascade, boundary conditions (including inlet velocity, pressure, temperature, etc.) under the target working condition, and inlet flow field data (such as velocity, pressure, density, etc. of each part of the flow field) at the inlet of the target cascade.

[0070] The working condition can be input into the flow field simulation software, and the steady flow field finally formed in the target cascade under the given working condition can be calculated by the flow field simulation software. This flow field is also the simulation flow field data. The flow field simulation software used in the embodiments of the present application is not limited, and any flow field simulation software is applicable to the embodiments of the present application.

[0071] At step 106, the working condition is subjected to feature extraction processing by the Fourier layer of the original flow field prediction model to obtain flow field feature data, and the flow field feature data is subjected to flow field reconstruction processing by the output layer of the original flow field prediction model to obtain predicted flow field data.

[0072] In the embodiments of the present application, the flow field prediction model is constructed based on a Fourier neural operator and includes a Fourier layer and an output layer. The Fourier layer can be multiple, and each Fourier layer is sequentially connected. The Fourier neural operator is a neural network model that extracts features of input data in the frequency domain by using Fourier transform. Since it is easier to capture global features and long-term dependencies of input data in the frequency domain, and the global features of flow fields under different working conditions have certain similarity, the flow field prediction model is constructed by using the Fourier neural operator, which can improve the performance of the model when migrated to other working conditions and speed up the fine-tuning training of the model under other working conditions.

[0073] Each Fourier layer includes a Fourier transform layer and an inverse Fourier transform layer. The Fourier transform layer is used to convert the working condition to the frequency domain and extract the frequency domain features by Fourier transform, and the inverse Fourier transform layer is used to convert the extracted flow field feature data back to the spatial domain. The output layer is used to reconstruct the flow field according to the flow field feature data extracted by the last Fourier layer to obtain predicted flow field data.

[0074] The working condition can be subjected to encoding and normalization processing before being input into the flow field prediction model. The size of the inlet flow field data is , where H and W are the resolutions of the inlet flow field data in the height and width directions, respectively, and C is the number of channels of the inlet flow field data, each channel representing a parameter of the inlet flow field data, such as velocity, pressure, etc. By means of spatial replication, the cascade parameters, boundary conditions, etc. can be expanded into additional channels with the same resolution as the inlet flow field data. Splicing each additional channel with the inlet flow field data can obtain encoded data with a size of , where C' is the number of additional channels. After further maximum and minimum normalization of the data of each channel, the input data input into the flow field prediction model can be obtained.

[0075] In one embodiment, as shown in Figure 2 , the step of extracting flow field feature data by the Fourier layer includes:

[0076] At step 202, the working condition is converted to the frequency domain, and a preset number of low-frequency components with a frequency less than a preset threshold are extracted from the working condition converted to the frequency domain.

[0077] At step 204, each low-frequency component is subjected to frequency domain feature modulation processing.

[0078] Step 206, convert each low-frequency component after frequency domain feature modulation to the spatial domain to obtain flow field feature data.

[0079] In the embodiment of the present application, after converting the working condition to the frequency domain, each low-frequency component of the working condition in the frequency domain can be obtained. Among these low-frequency components, a preset number of low-frequency components with a frequency less than a preset threshold can be retained for frequency domain feature extraction. The preset number and the preset threshold can both be set by a person skilled in the art according to requirements, for example, the preset number can be set to half of the resolution in the height of the working condition data; or the preset number and the preset threshold can also be adjusted according to the prediction effect during model training. The reason for using low-frequency components for feature extraction is that low-frequency components often reflect the global features of the flow field and carry the main information required for subsequent flow field reconstruction, so filtering out high-frequency components and using low-frequency components for feature extraction can reduce the amount of calculation while maintaining the accuracy of flow field reconstruction.

[0080] After filtering to obtain each low-frequency component, each low-frequency component is multiplied by a learnable parameter matrix to perform frequency domain feature modulation processing on each low-frequency component. The learnable parameter matrix is a complex domain matrix of size k x k (k is the number of low-frequency components), and the elements in the matrix can be adjusted according to the prediction effect during training.

[0081] After frequency domain feature modulation, each low-frequency component obtained is converted to the spatial domain by inverse Fourier transform, and flow field feature data output by the Fourier layer can be obtained.

[0082] Step 108, determine a loss value according to the difference between the predicted flow field data and the simulation flow field data, adjust the output layer of the original flow field prediction model based on the loss value, and obtain a trained flow field prediction model corresponding to the target working condition.

[0083] In the embodiment of the present application, after obtaining the predicted flow field data output by the output layer, the loss value of the model can be calculated according to the difference between the predicted flow field data and the simulation flow field data, and the model can be adjusted according to the loss value. When fine-tuning the original flow field prediction model, the parameters of the Fourier layer of the original flow field prediction model can be frozen, and only the parameters of the output layer can be adjusted, so as to retain the global feature extraction capability of the Fourier layer of the original flow field prediction model trained, and at the same time enable the original flow field prediction model to learn the local features of the target working condition, and enable the original flow field prediction model to reconstruct the predicted flow field data that conforms to the characteristics of the target working condition according to the flow field feature data extracted by the Fourier layer.

[0084] The embodiments of the present application do not make specific limitations on the calculation manner of the loss value calculated according to the difference between the predicted flow field data and the simulated flow field data, and any loss function is applicable to the embodiments of the present application. The training can be stopped when the number of training iterations reaches a preset number or the loss is less than a preset value, and the original flow field prediction model at this time is taken as the flow field prediction model corresponding to the target working condition.

[0085] The flow field prediction model training method based on the Fourier neural operator provided by the embodiments of the present application applies the Fourier neural operator to flow field prediction, and divides the flow field prediction model into a Fourier layer adopting the Fourier neural operator architecture and an output layer. When it is necessary to train the flow field prediction model for a target working condition, an original flow field prediction model trained under other working conditions is obtained, the output layer of the original flow field prediction model is fine-tuned under the working condition of the target working condition, and a flow field training model of the target working condition is obtained. Since it is easier to capture the global characteristics and long-term dependencies of the flow field in the frequency domain, and the global characteristics of the flow field under different working conditions have certain similarity, the Fourier neural operator is used to construct the flow field prediction model, and when the model is migrated to other working conditions, the parameters of the Fourier layer do not need to be adjusted, and the parameters of the output layer can be adjusted to obtain a flow field reconstruction model conforming to the characteristics of the target working condition, which can speed up the fine-tuning training of the model under other working conditions and improve the training efficiency of the flow field prediction model.

[0086] In one embodiment, the original flow field prediction model is trained based on an initial flow field prediction model, and the initial flow field prediction model adjusts the Fourier layer and the output layer based on the loss value during the training process.

[0087] In the embodiments of the present application, the original flow field prediction model is trained based on an initial flow field prediction model, and the process of training the original flow field prediction model can be referred to the process of training the flow field prediction model based on the original flow field prediction model described above, which will not be repeated here. The difference between the two training processes is that only the output layer parameters are adjusted according to the loss value in the process of training the flow field prediction model based on the original flow field prediction model, but the parameters of the Fourier layer and the parameters of the output layer are adjusted according to the loss value in the process of training the original flow prediction model based on the initial flow field prediction model. The preset number, the preset threshold, and each element in the learnable parameter matrix in the foregoing embodiments can be adjusted in the process of training the original flow prediction model based on the initial flow field prediction model.

[0088] In one embodiment, as shown in FIG. 1, Figure 3 In step 108, the loss value is determined according to the difference between the predicted flow field data and the simulated flow field data, including:

[0089] In step 302, the physical residual loss of the predicted flow field data is determined based on the Navier-Stokes equation.

[0090] Step 304, determining a flow field prediction loss based on a difference between the predicted flow field data and the simulated flow field data;

[0091] Step 306, determining a loss value based on the physical residual loss and the flow field prediction loss.

[0092] In the embodiments of the present application, the physical residual loss is also considered in the process of calculating the loss value, so that the predicted flow field data obtained by reconstruction is more in line with the physical law. The Navier-Stokes equation is used to describe the basic law of fluid motion, and the physical residual loss of the predicted flow field data on the Navier-Stokes equation can be obtained by calculating the physical residual loss of the predicted flow field data on the continuity equation and the momentum conservation equation respectively and combining the two physical residual losses.

[0093] The continuity equation represents that the fluid is continuous. When the continuity equation represents mass conservation, the meaning of the equation is that the rate of change of fluid density with time at a point in the flow field should be equal to the mass flux of the fluid flowing out of the unit volume per unit volume at the point. The physical residual loss of the predicted flow field data on the continuity equation can be the sum of the rate of change of fluid density with time and the divergence of mass flux (see formula (I)), which should be 0 in an ideal case:

[0094] Formula (I)

[0095] Wherein is the physical residual loss on the continuity equation, is the density, is the velocity vector.

[0096] The momentum conservation equation represents that the rate of change of momentum of the fluid with time should be equal to the sum of the external forces received by the fluid. The physical residual loss of the predicted flow field data on the momentum conservation equation can be the difference between the rate of change of momentum of the fluid with time and the external forces received by the fluid (see formula (II)), which should be 0 in an ideal case:

[0097] Formula (II)

[0098] Wherein is the physical residual loss on the momentum conservation equation, represents the change of momentum of the fluid, is the pressure, is the viscous stress tensor (T , is the dynamic viscosity, T is the transpose, is the unit tensor), is the gravitational acceleration.

[0099] The physical residual loss of the two is obtained by adding the two-norm of the two, and the physical residual loss of the predicted flow field data on the Navier-Stokes equation is obtained, see equation (three):

[0100] Equation (three)

[0101] Wherein, is the physical residual loss, N is the total number of training samples in this batch, and i represents the i th sample.

[0102] The flow field prediction loss can be determined based on the difference between the predicted flow field data and the simulation flow field data at each point. The loss value is obtained by weighted sum of the physical residual loss and the flow field prediction loss.

[0103] In one embodiment, the above method further comprises:

[0104] The total pressure loss prediction processing is performed on the flow field feature data through the output layer of the original flow field prediction model, and predicted total pressure loss data is obtained.

[0105] In step 306, the loss value is determined based on the physical residual loss and the flow field prediction loss, comprising:

[0106] The total pressure loss prediction loss is determined according to the difference between the predicted total pressure loss data and the simulation total pressure loss data.

[0107] The loss value is determined based on the physical residual loss, the flow field prediction loss and the total pressure loss prediction loss.

[0108] In the embodiments of the present application, the output layer can also predict the total pressure loss according to the flow field feature data. The total pressure loss represents the mechanical energy loss of the fluid due to friction, separation and other reasons during movement, and can be used to measure the aerodynamic performance of the target cascade. The output layer outputs the predicted total pressure loss in addition to the predicted flow field data, which facilitates the evaluation of the performance of the target cascade.

[0109] In the case where the output layer predicts the total pressure loss, the total pressure loss prediction loss can be considered in addition to the physical residual loss and the flow field prediction loss when calculating the loss value. The total pressure loss prediction loss can be calculated by the difference between the predicted total pressure loss data and the simulation total pressure loss data, wherein the simulation total pressure loss data can be calculated according to the pressure difference of each position in the simulation flow field data. The physical residual loss, the flow field prediction loss and the total pressure loss prediction loss can be respectively set with weights, and the final loss value is obtained by weighted sum of the above three losses (see equation (four)):

[0110] Equation (four)

[0111] Wherein, is the loss value, to predict the flow field data, to simulate the flow field data, refers to calculating the difference between the predicted flow field data and the simulated flow field data by mean square error. to predict the total pressure loss, to simulate the total pressure loss, refers to calculating the difference between the predicted total pressure loss and the simulated total pressure loss by mean absolute error. to predict the physical residual loss, 、 、 respectively, which can be 0.8, 0.2, 0.5 respectively.

[0112] In one embodiment, as shown in Figure 4 a flow field prediction method based on a Fourier neural operator is provided, the method comprising:

[0113] Step 402, determining the working condition of the target working condition;

[0114] Step 404, inputting the working condition into the flow field prediction model to obtain the predicted flow field data corresponding to the target cascade under the target working condition;

[0115] The flow field prediction model is trained by the method of any of the preceding embodiments.

[0116] In the embodiments of the present application, in actual application, the working condition of the flow field generated in the target cascade under the target working condition can be obtained, and then the working condition is input into the flow field prediction model trained in the manner of the preceding embodiments to obtain the predicted flow field data of the target cascade. The flow field prediction model can also output the total pressure loss. After obtaining the predicted flow field data and the total pressure loss, the server can further draw the flow field cloud map and the total pressure loss trend curve, and display the pressure distribution in the flow field and the total pressure loss trend in the flow field in a visual form.

[0117] It should be understood that although each step in the flowchart involved in each embodiment as described above is displayed in sequence according to the arrow, these steps are not necessarily executed in sequence according to the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other sequences. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution sequence of these steps or stages is not necessarily sequential, but can be executed alternately or alternately with at least part of other steps or steps or stages in other steps.

[0118] Based on the same inventive concept, the application further provides a Fourier neural operator based flow field prediction model training device for implementing the Fourier neural operator based flow field prediction model training method described above. The implementation scheme for solving problems provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more Fourier neural operator based flow field prediction model training device embodiments provided below can be referred to the limitations of the Fourier neural operator based flow field prediction model training method described above, which will not be repeated here.

[0119] In one embodiment, as shown in Figure 5 A Fourier neural operator based flow field prediction model training device 500 is provided, the flow field prediction model is constructed based on a Fourier neural operator and includes a Fourier layer and an output layer, and the device includes a determination module 502, a generation module 504, a reconstruction module 506, and a training module 508, wherein:

[0120] The determination module 502 is configured to determine an original flow field prediction model corresponding to a target working condition, wherein the original flow field prediction model is trained based on a working condition of an original working condition;

[0121] The generation module 504 is configured to generate simulation flow field data corresponding to a target cascade under the target working condition according to a working condition of the target working condition;

[0122] The reconstruction module 506 is configured to perform feature extraction processing on the working condition through the Fourier layer of the original flow field prediction model to obtain flow field feature data, and perform flow field reconstruction processing on the flow field feature data through the output layer of the original flow field prediction model to obtain predicted flow field data;

[0123] The training module 508 is configured to determine a loss value according to a difference between the predicted flow field data and the simulation flow field data, adjust the output layer of the original flow field prediction model based on the loss value, and obtain a trained flow field prediction model corresponding to the target working condition.

[0124] The device for training a flow field prediction model based on a Fourier neural operator provided by the embodiment of the application applies the Fourier neural operator to flow field prediction, and divides the flow field prediction model into a Fourier layer adopting a Fourier neural operator architecture and an output layer. When it is necessary to train a flow field prediction model for a target working condition, an original flow field prediction model trained under other working conditions is acquired, and the output layer of the original flow field prediction model is fine-tuned by using the working condition of the target working condition to obtain a flow field training model of the target working condition. Since it is easier to capture global features and long-term dependencies of a flow field in the frequency domain, and the global features of the flow field under different working conditions have certain similarity, the Fourier neural operator is used to construct the flow field prediction model, and when the model is migrated to other working conditions, the parameters of the Fourier layer do not need to be adjusted, and the parameters of the output layer can be adjusted to obtain a flow field reconstruction model conforming to the characteristics of the target working condition, so that the speed of fine-tuning and training the model under other working conditions can be accelerated, and the training efficiency of the flow field prediction model can be improved.

[0125] In one of the embodiments, the original flow field prediction model is trained based on an initial flow field prediction model, and the initial flow field prediction model adjusts the Fourier layer and the output layer based on a loss value during the training process.

[0126] In one of the embodiments, the reconstruction module 506 is further configured to:

[0127] convert the working condition to the frequency domain, and extract a preset number of low-frequency components with a frequency less than a preset threshold from the working condition converted to the frequency domain;

[0128] perform frequency domain feature modulation processing on each of the low-frequency components;

[0129] convert each of the low-frequency components after the frequency domain feature modulation to the spatial domain to obtain flow field feature data.

[0130] In one of the embodiments, the training module 508 is further configured to:

[0131] determine a physical residual loss of the predicted flow field data based on the Navier-Stokes equation;

[0132] determine a flow field prediction loss based on the difference between the predicted flow field data and the simulated flow field data;

[0133] determine a loss value based on the physical residual loss and the flow field prediction loss.

[0134] In one of the embodiments, the reconstruction module 506 is further configured to:

[0135] perform total pressure loss prediction processing on the flow field feature data through the output layer of the original flow field prediction model to obtain predicted total pressure loss data;

[0136] The training module 508 is further configured to:

[0137] determine a total pressure loss prediction loss according to a difference between the predicted total pressure loss data and the simulated total pressure loss data;

[0138] determine a loss value based on the physical residual loss, the flow field prediction loss and the total pressure loss prediction loss.

[0139] In one embodiment, as shown in Figure 6 A flow field prediction device 600 based on a Fourier neural operator is provided, the device comprising a determination module 602 and an input module 604, wherein:

[0140] The determination module 602 is configured to determine a working condition of a target working condition;

[0141] The input module 604 is configured to input the working condition to a flow field prediction model to obtain predicted flow field data corresponding to a target blade row under the target working condition;

[0142] The flow field prediction model is trained by the method of any of the preceding embodiments.

[0143] Each module in the device can be realized wholly or partially by software, hardware and a combination thereof. Each module can be embedded in or independent of a processor in a computer device in a hardware form, or stored in a memory in a computer device in a software form, so as to be called and executed by a processor to perform operations corresponding to each module.

[0144] In one embodiment, a computer device is provided, which can be a server, and an internal structure diagram of the computer device can be as shown in Figure 7 The computer device comprises a processor, a memory and a network interface connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for running the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement a flow field prediction model training method based on a Fourier neural operator.

[0145] Those skilled in the art can understand, Figure 7The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0146] In an embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor implementing the steps in the above method embodiments when executing the computer program.

[0147] In an embodiment, a computer readable storage medium is provided, storing a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.

[0148] In an embodiment, a computer program product is provided, including a computer program, and the computer program implementing the steps in the above method embodiments when executed by a processor.

[0149] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties.

[0150] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when the computer program is executed, the processes of the above-mentioned embodiments of the methods can be included. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0151] Any combination of the technical features of the above embodiments can be made. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist, it should be considered as the scope of the present application.

[0152] The above embodiments only express several implementation manners of the present application, and the description is more specific and detailed, but it should not be understood as a limitation on the scope of the patent of the present application. It should be pointed out that for ordinary skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are within the scope of protection of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A method for training a flow field prediction model based on a Fourier neural operator, characterized in that, The flow field prediction model is constructed based on a Fourier neural operator and includes a Fourier layer and an output layer, and the method includes: determining an original flow field prediction model corresponding to a target working condition, the original flow field prediction model being trained based on working condition conditions of an original working condition; generating simulation flow field data corresponding to a target blade row in the target working condition according to working condition conditions of the target working condition; performing feature extraction processing on the working condition conditions through the Fourier layer of the original flow field prediction model to obtain flow field feature data, and performing flow field reconstruction processing on the flow field feature data through the output layer of the original flow field prediction model to obtain predicted flow field data; determining a loss value according to a difference between the predicted flow field data and the simulation flow field data, adjusting the output layer of the original flow field prediction model based on the loss value to obtain a trained flow field prediction model corresponding to the target working condition.

2. The method of claim 1, wherein, The original flow field prediction model is trained based on an initial flow field prediction model, and the initial flow field prediction model adjusts the Fourier layer and the output layer based on a loss value during training.

3. The method of claim 1, wherein, The feature extraction processing on the working condition conditions through the Fourier layer of the original flow field prediction model to obtain flow field feature data includes: converting the working condition conditions to a frequency domain, and extracting a preset number of low-frequency components with a frequency less than a preset threshold from the working condition conditions converted to the frequency domain; performing frequency domain feature modulation processing on each low-frequency component; converting each low-frequency component after the frequency domain feature modulation to a spatial domain to obtain flow field feature data.

4. The method of claim 1, wherein, The determination of the loss value according to the difference between the predicted flow field data and the simulation flow field data includes: determining a physical residual loss of the predicted flow field data based on the Navier-Stokes equation; determining a flow field prediction loss based on the difference between the predicted flow field data and the simulation flow field data; determining a loss value based on the physical residual loss and the flow field prediction loss.

5. The method of claim 4, wherein, The method further includes: performing total pressure loss prediction processing on the flow field feature data through the output layer of the original flow field prediction model to obtain predicted total pressure loss data; The determination of the loss value based on the physical residual loss and the flow field prediction loss includes: determining a total pressure loss prediction loss according to a difference between the predicted total pressure loss data and simulation total pressure loss data; determining a loss value based on the physical residual loss, the flow field prediction loss, and the total pressure loss prediction loss.

6. A method for flow field prediction based on Fourier Neural Operator, characterized in that, The method includes: determining working condition conditions of a target working condition; inputting the working condition conditions to a flow field prediction model to obtain predicted flow field data corresponding to a target blade row in the target working condition; wherein the flow field prediction model is trained by the method of any one of claims 1-5. 7.A device for training a flow field prediction model based on a Fourier neural operator, characterized in that, The flow field prediction model is constructed based on a Fourier neural operator and includes a Fourier layer and an output layer, and the device includes: a determination module configured to determine an original flow field prediction model corresponding to a target working condition, the original flow field prediction model being trained based on working condition conditions of an original working condition; The generating module is configured to generate simulation flow field data corresponding to a target cascade under the target working condition according to a working condition of the target working condition; The reconstructing module is configured to perform feature extraction processing on the working condition by a Fourier layer of the original flow field prediction model to obtain flow field feature data, and perform flow field reconstruction processing on the flow field feature data by an output layer of the original flow field prediction model to obtain predicted flow field data; The training module is configured to determine a loss value according to a difference between the predicted flow field data and the simulation flow field data, adjust the output layer of the original flow field prediction model based on the loss value, and obtain a trained flow field prediction model corresponding to the target working condition.

8. A flow field prediction device based on Fourier Neural Operator, characterized by, The device comprises: A determining module configured to determine a working condition of a target working condition; An input module configured to input the working condition to a flow field prediction model to obtain predicted flow field data corresponding to a target cascade under the target working condition; The flow field prediction model is trained by the method of any one of claims 1-5. 9.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is configured to perform the method according to any one of claims 1-8 when the computer program is executed by the processor. The processor executes the computer program to implement the steps of the method of any one of claims 1-6.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1-6.

Images