Flow field prediction method, device, terminal equipment and storage medium

Abstract

The present application is applicable to the field of flow field prediction technology, and provides a flow field prediction method, apparatus, terminal device, and storage medium, including: obtaining a first coordinate corresponding to a point to be measured, where the point to be measured is a coordinate point in a calculation domain; using the first coordinate as the input of a pre-trained neural network model to obtain first flow field data output by the neural network model, wherein the first flow field data reflects the air flow condition corresponding to the point to be measured; wherein the neural network model reconstructs the flow field in the calculation domain based on the physical laws described by the fluid control equation, predicts the air flow condition of the point to be measured based on the flow field and the first coordinate, and obtains the first flow field data. The present application can improve the accuracy of flow field prediction.

Description

Technical Field

[0001] The present application belongs to the field of flow field prediction technology, and in particular relates to a flow field prediction method, apparatus, terminal device, and computer-readable storage medium. Background Art

[0002] The flow field refers to the distribution of flow velocity and wind direction within a certain area, which has a direct impact on a city's climate and pollutant dispersion. In practical engineering applications, the study of flow fields is of great significance to applications such as architectural design and urban climate analysis.

[0003] At present, computational fluid dynamics (CFD) simulation is usually used to solve the differential equations of fluid flow to simulate the air flow conditions in the flow field. However, CFD simulation requires the construction of a large number of grids, and the accuracy is affected by factors such as grid resolution. The calculation is complex and the accuracy is low, which is not conducive to practical application. Summary of the Invention

[0004] The embodiments of the present application provide a flow field prediction method, apparatus, terminal device, and storage medium, which can improve the accuracy of flow field prediction.

[0005] In a first aspect, an embodiment of the present application provides a flow field prediction method, comprising:

[0006] Obtaining a first coordinate corresponding to a point to be measured, where the point to be measured is a coordinate point in the calculation domain;

[0007] Using the first coordinate as an input of a pre-trained neural network model to obtain first flow field data output by the neural network model, wherein the first flow field data reflects the air flow condition corresponding to the point to be measured;

[0008] The neural network model reconstructs the flow field in the calculation domain based on the physical laws described by the fluid control equation, predicts the air flow condition of the measured point according to the flow field and the first coordinate, and obtains the first flow field data.

[0009] In a second aspect, an embodiment of the present application provides a flow field prediction device, comprising:

[0010] A first coordinate acquisition module is used to acquire a first coordinate corresponding to a point to be measured, where the point to be measured is a coordinate point in the calculation domain;

[0011] a prediction module, configured to use the first coordinate as an input of a pre-trained neural network model to obtain first flow field data output by the neural network model, wherein the first flow field data reflects the air flow condition corresponding to the point to be measured;

[0012] The neural network model reconstructs the flow field in the calculation domain based on the physical laws described by the fluid control equation, predicts the air flow condition of the measured point according to the flow field and the first coordinate, and obtains the first flow field data.

[0013] In a third aspect, an embodiment of the present application provides a terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the flow field prediction method described in the first aspect when executing the computer program.

[0014] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the flow field prediction method described in the first aspect are implemented.

[0015] In a fifth aspect, an embodiment of the present application provides a computer program product, which, when executed on a terminal device, enables the terminal device to execute the flow field prediction method described in any one of the above-mentioned first aspects.

[0016] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0017] In the embodiment of the present application, since the neural network model can reconstruct the flow field in the computational domain based on the physical laws described by the fluid control equations, that is, the physical information described by the fluid control equations is embedded in the neural network model, the pre-trained neural network model can solve the fluid control equations and reconstruct the flow field in the computational domain from the perspective of the physical laws described by the fluid control equations. Compared with solving the fluid control equations through CFD simulation, there is no need to construct a large number of grids. While reducing the complexity of flow field reconstruction, the accuracy of the reconstructed flow field is not affected by the grid resolution. Therefore, the flow field in the computational domain is reconstructed by the neural network model embedded with the fluid control equations, and the air flow conditions of the test point are queried based on the reconstructed flow field and the first coordinate to obtain the required first flow field data. While reducing the computational cost of flow field reconstruction and prediction, it is not affected by the grid resolution, thereby improving the accuracy of flow field prediction and facilitating practical engineering applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art.

[0019] Figure 1 This is a flow chart of a flow field prediction method provided in one embodiment of the present application;

[0020] Figure 2 This is a schematic diagram of a coordinate system structure provided in an embodiment of the present application;

[0021] Figure 3 Schematic diagram of the structure of the flow field prediction device provided in the embodiment of the present application;

[0022] Figure 4 It is a structural diagram of the terminal device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In the following description, specific details such as specific system structures and techniques are provided for purposes of illustration rather than limitation to facilitate a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obscuring the description of the present application with unnecessary detail.

[0024] It should be understood that when used in the present specification and the appended claims, the term "comprising" indicates the presence of described features, integers, steps, operations, elements and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0025] It will also be understood that the term "and / or" used in this specification and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0026] In addition, in the description of the present application specification and the appended claims, the terms "first", "second", "third", etc. are only used to distinguish the descriptions and cannot be understood as indicating or implying relative importance.

[0027] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with the embodiment is included in one or more embodiments of the present application. Thus, phrases such as "in one embodiment," "in some embodiments," "in other embodiments," and "in yet other embodiments" appearing in various places in this specification do not necessarily refer to the same embodiment, but rather mean "one or more but not all embodiments," unless otherwise specifically emphasized.

[0028] Example 1:

[0029] Figure 1 A flow chart of a flow field prediction method provided by an embodiment of the present invention is shown, and is described in detail as follows:

[0030] Step S101: obtaining a first coordinate corresponding to a point to be measured, where the point to be measured is a coordinate point in a calculation domain.

[0031] The above-mentioned computational domain refers to the area used to analyze DC flow and heat transfer calculations. When reconstructing the flow field, it is usually necessary to determine a specific computational domain (i.e., the scope of the computational domain) and the various boundary conditions corresponding to the computational domain to reconstruct the flow field within the computational domain.

[0032] Specifically, when it is necessary to predict the air flow condition of a certain position (ie, a point to be measured) in the calculation domain, the spatial coordinates of the point to be measured are obtained to obtain the first coordinates.

[0033] In step S102 , the first coordinate is used as an input of a pre-trained neural network model to obtain first flow field data output by the neural network model. The first flow field data reflects the air flow condition corresponding to the point to be measured.

[0034] The neural network model reconstructs the flow field in the calculation domain based on the physical laws described by the fluid control equations, predicts the air flow condition of the test point according to the flow field and the first coordinates, and obtains the first flow field data.

[0035] Optionally, the fluid control equations may be control equations such as the Reynolds equations or the Euler equations. In some embodiments, since the time-dependent fluid control equations require a large amount of computation and are difficult to solve, there are currently no effective solutions for solving the Navier-Stokes equations at high Reynolds numbers and using the Navier-Stokes equations to solve engineering problems related to turbulence. Therefore, in the embodiments of the present application, the fluid control equations may be variants of the Navier-Stokes equations, namely the Reynolds time-averaged equations or other time-averaged fluid control equations, i.e., the fluid control equations are steady-state, time-independent fluid control equations, and the resulting solutions are also steady-state solutions, such as time-averaged pressure, so as to reduce the difficulty and computational complexity of solving the fluid control equations while ensuring the accuracy of the flow field prediction.

[0036] The above-mentioned flow field refers to the spatial region occupied by the movement of fluid (including gas and liquid), that is, the spatial distribution of fluid flow.

[0037] Specifically, since the neural network model can reconstruct the flow field within the calculation domain based on the physical laws described by the fluid control equations, when it is necessary to predict the air flow conditions of the test point within the calculation domain, the first coordinate corresponding to the test point can be used as the input of the neural network model. The neural network model queries the air flow conditions of the test point corresponding to the first coordinate in the reconstructed flow field to obtain the first flow field data.

[0038] It can be understood that the air flow condition at the above-mentioned test point is determined by the fluid (such as liquid and / or gas) in the calculation domain, and the above-mentioned first flow field data may include the flow velocity and / or pressure of the fluid at the test point, wherein the flow velocity may be the combined velocity of the fluid at the test point, or the flow velocity of the test point in different spatial directions, that is, the first flow field data may include the flow velocity components of the fluid in different spatial directions. For example, in the calculation domain, Figure 2 In the coordinate system shown, the first flow field data may include flow velocities of the fluid in the x-direction, the y-direction, and the z-direction.

[0039] In the embodiments of the present application, since the fluid governing equations are embedded in the neural network model, the pre-trained neural network model can solve the embedded fluid governing equations and reconstruct the flow field within the computational domain based on the physical laws described by the fluid governing equations. Compared to solving the fluid governing equations through CFD simulation, this method does not require the construction of a large number of grids and is therefore not affected by grid resolution. Therefore, the pre-trained neural network model embedded with the fluid governing equations can quickly and accurately reconstruct the flow field within the computational domain and query the air flow conditions at the test point based on the reconstructed flow field and the first coordinate to obtain the required first flow field data. Furthermore, since the physical information constraints are embedded in the neural network model, the pre-trained neural network model can be obtained based on a small amount of measured data, compared to a purely data-driven neural network model. Furthermore, compared to fluid governing equations that do not consider the effects of temperature, humidity, and other factors in the actual environment, the neural network model embedded with the fluid governing equations is trained based on measured data and can reconstruct and predict the flow field in the computational domain based on both the physical laws described by the fluid governing equations and the actual conditions of the computational domain. This can improve the accuracy of flow field prediction and is beneficial for practical engineering applications.

[0040] In some embodiments, the fluid control equation is the Reynolds time-averaged equation, and the pre-trained neural network model is obtained according to the following steps:

[0041] A1. Determine the Reynolds time-averaged residual function corresponding to the above-mentioned calculation domain based on the above-mentioned Reynolds time-averaged equation.

[0042] A2. Determine the loss function of the neural network model to be trained based on the Reynolds time-averaged residual function to obtain the neural network model to be trained.

[0043] A3. Obtain training data, where the training data includes a plurality of second flow field data collected from sampling points in the calculation domain.

[0044] A4. Train the neural network model to be trained according to the training data until the trained neural network model meets the preset requirements, thereby obtaining the pre-trained neural network model.

[0045] Specifically, since the Reynolds time-averaged equation can better describe the flow characteristics of the fluid, in the embodiment of the present application, the Reynolds time-averaged equation is used as the physical equation embedded in the neural network model, the Reynolds time-averaged residual function corresponding to the calculation domain is determined according to the Reynolds time-averaged equation, and the loss function of the neural network model to be trained is determined according to the Reynolds time-averaged residual function, that is, in the process of training the neural network model to be trained using training data, the error between the predicted value and the measured value (i.e., the second flow field data) of the neural network model is determined in combination with the physical law described by the Reynolds time-averaged equation, thereby improving the accuracy of the trained neural network model. In the training process of the neural network model, if the trained neural network model meets the preset requirements (such as the accuracy reaches a preset threshold of 0.98, and / or the number of iterations reaches a preset number of 1000, etc.), the training of the neural network model can be stopped to obtain a pre-trained neural network model.

[0046] In the embodiment of the present application, the Reynolds time-averaged equation is used as the fluid control equation embedded in the neural network model, and the Reynolds time-averaged equation can better describe the flow characteristics of fluids such as gases, so that the neural network model embedded with the Reynolds time-averaged equation can better reconstruct the flow field based on the fluid flow characteristics described by the Reynolds time-averaged equation, thereby improving the prediction accuracy of the neural network model. In addition, since physical information constraints are embedded in the neural network model, compared to a purely data-driven neural network model, a pre-trained neural network model can be obtained based on a small amount of measured data. At the same time, compared to the fluid control equation that does not consider the influence of temperature, humidity, etc. in the actual environment, the neural network model embedded with the fluid control equation is trained based on measured data, and can simultaneously reconstruct the flow field of the computational domain based on the physical laws described by the fluid control equation and the actual situation of the computational domain for prediction, which can improve the prediction accuracy of the trained neural network model and is beneficial to actual engineering applications.

[0047] In some embodiments, step A1 includes:

[0048] A11. Determine the momentum residual functions corresponding to multiple spatial directions according to the above Reynolds time-averaged equation.

[0049] A12. Determine the Reynolds time-averaged residual function according to the momentum residual function corresponding to each of the spatial directions.

[0050] Specifically, since the flow of fluids such as air exists in multiple spatial directions such as the vertical direction, in order to more accurately reconstruct the flow field in the calculation domain and improve the prediction accuracy of the air flow conditions in the calculation domain, in the embodiment of the present application, based on automatic differentiation, at least two spatial directions (such as Figure 2The momentum residual function corresponding to the x-direction and z-direction shown in FIG is constructed, and then the overall Reynolds time-averaged residual function is determined based on the momentum residual functions of multiple spatial directions.

[0051] In some embodiments, a continuity equation residual function combining multiple spatial directions can also be constructed based on automatic differentiation and Reynolds time-averaged equations to fully combine the impact of the overall fluidity of the air fluid on the air flow conditions and improve the prediction accuracy.

[0052] In an embodiment of the present application, since the influence of air flow on the flow field in the calculation domain is considered from multiple spatial directions when determining the Reynolds time-averaged residual function, the momentum residual functions of multiple spatial directions are determined according to the Reynolds time-averaged equation, and then the final Reynolds time-averaged residual function is comprehensively determined based on the momentum residual functions of multiple spatial directions. Therefore, determining the loss function of the neural network model based on the Reynolds residual function can improve the prediction accuracy of the neural network model.

[0053] In some embodiments, before the above step A11, the method further includes:

[0054] A10. Determine the closed Reynolds time-averaged equation based on the Reynolds time-averaged equation and the zero equation.

[0055] Correspondingly, the above step A11 includes:

[0056] The momentum residual function corresponding to each of the spatial directions is determined respectively according to the closed Reynolds time-averaged equation.

[0057] The zero equation mentioned above refers to an equation that uses an algebraic relationship to relate the eddy viscosity coefficient to its time-averaged value.

[0058] Specifically, in order to facilitate the solution of the Reynolds time-averaged equation, reduce the computational difficulty and improve the accuracy of the neural network model, in an embodiment of the present application, a zero equation (such as Li Cheng's zero equation) is first used to close the Reynolds time-averaged equation, and then, the momentum residual function corresponding to each spatial direction can be better determined based on the closed Reynolds time-averaged equation.

[0059] In some embodiments, the above Reynolds time-averaged equation can be expressed as:

[0060]

[0061]

[0062] Where ρ is the fluid density, and the above μ eff represents the equivalent dynamic viscosity, t represents time, i represents the spatial direction, including x, y and z directions, V i Including the flow velocity in each spatial direction, V x(i=x) is the flow velocity u in the x direction, V y (i=y) represents the flow velocity v in the y direction, V z (i=z) represents the flow velocity w in the z direction. Similarly, j also represents the spatial direction, including the x direction, y direction and z direction.

[0063] The above equivalent dynamic viscosity can be expressed as follows:

[0064] μ eff =μ t +μ

[0065] Among them, μ t is the turbulent dynamic viscosity, and μ is the laminar dynamic viscosity.

[0066] Optionally, the third flow field data includes the flow velocity u in the x-direction, the flow velocity v in the y-direction, and the flow velocity w in the z-direction of the sampling point. The momentum residual functions in multiple spatial directions determined according to the closed Reynolds time-averaged equation can be in the following form:

[0067]

[0068]

[0069]

[0070] Among them, f1, f2 and f3 correspond to the momentum residual functions in the x-direction, y-direction and z-direction respectively, u is the flow velocity in the x-direction of the third flow field data, v is the flow velocity in the y-direction of the third flow field data, w is the flow velocity in the z-direction of the third flow field data, ρ is the density of the fluid (such as air) at the sampling point corresponding to the third flow field data, t represents time, μ represents laminar dynamic viscosity, μ t represents the turbulent dynamic viscosity.

[0071] In some embodiments, when determining the momentum residual functions in multiple spatial directions according to the closed Reynolds time-averaged equation, a residual function of the continuity equation can also be determined according to the closed Reynolds time-averaged equation. The residual function of the continuity equation can be expressed as follows:

[0072]

[0073] In an embodiment of the present application, the zero equation is embedded into the Reynolds time-averaged equation to close the turbulent dynamic viscosity and obtain a closed Reynolds time-averaged equation, which facilitates the solution of the Reynolds time-averaged control equation, and further facilitates the determination of the momentum residual function in multiple spatial directions. When the neural network model is trained based on the predicted third flow field data and the measured second flow field data, the error of the predicted third flow field data can be analyzed separately from multiple spatial directions, so as to optimize the neural network model based on the error, thereby improving the accuracy of the neural network model and reducing the training difficulty.

[0074] In some embodiments, step A10 includes:

[0075] The Li Chengling equation is embedded in the above-mentioned Reynolds time-averaged equation to obtain the closed Reynolds time-averaged equation, wherein the turbulent dynamic viscosity in the closed Reynolds time-averaged equation is determined according to the larger value of the turbulent dynamic viscosity near the wall and the turbulent dynamic viscosity at the far end in the above-mentioned calculation domain.

[0076] Specifically, in an embodiment of the present application, the Li Cheng zero equation model is embedded in the Reynolds time-averaged equation, and the unclosed turbulent stress and turbulent viscosity in the Reynolds time-averaged equation are combined through the Li Cheng zero equation, so that the unknown quantity is converted into a turbulent viscosity variable, so that the closed Reynolds time-averaged equation can be easily and accurately solved to obtain the turbulent dynamic viscosity in the computational domain. The turbulent dynamic viscosity is an indicator that describes the fluid dynamics and flow properties, which can reflect the intensity of the fluid drag force in the computational domain, and reflect the flow state and temperature field distribution of the fluid. Therefore, the Reynolds time-averaged equation closed by embedding the Li Cheng zero equation can better describe the fluid motion in the computational domain, thereby improving the accuracy of the flow field reconstructed by the neural network model embedded in the Reynolds time-averaged equation.

[0077] In some embodiments, the closed Reynolds time-averaged equation embedded in the Li Chengling equation can be expressed as follows:

[0078]

[0079] Where ρ is the fluid density, μ is the laminar dynamic viscosity, i represents the spatial direction, including the x-direction, y-direction and z-direction, V i Including the flow velocity in each spatial direction, V x (i=x) is the flow velocity u in the x direction, V y (i=y) represents the flow velocity v in the y direction, V z (i=z) represents the flow velocity w in the z direction. Similarly, j also represents the spatial direction, including the x direction, y direction and z direction.

[0080] In the embodiment of the present application, after the Li Chengling equation is embedded into the Reynolds time-averaged equation, the turbulent dynamic viscosity μ in the closed Reynolds time-averaged equation ist It is expressed as follows:

[0081] μ t =max(μ in , μ out )

[0082] Among them, μ in and μ out They represent the turbulent dynamic viscosity near the wall and at the far end of the computational domain, that is, after embedding Li Cheng's zero equation, the turbulent dynamic viscosity μ near the wall is taken in and the turbulent dynamic viscosity μ at the far end out The larger value of is used as the turbulent dynamic viscosity in the closed Reynolds time-averaged equation. The turbulent dynamic viscosity can reflect the strength of the fluid drag force in the computational domain. The larger of the two turbulent dynamic viscosities corresponding to the near wall and the far end is used as the solved turbulent dynamic viscosity. That is, the flow field in the computational domain is reconstructed by the fluid drag force with a larger intensity to improve the accuracy of the reconstructed flow field.

[0083] Optionally, the turbulent dynamic viscosity μ near the wall is in It can be expressed in the form of the Prandtl mixing length model as follows:

[0084] μ in =(C in l) 2 S

[0085] Among them, the parameter Cin can be expressed as follows:

[0086]

[0087] Where C represents the width of the windward side of the building in the computational domain, and H represents the height of the windward side of the building in the computational domain.

[0088] The parameter S can be expressed as follows:

[0089]

[0090] Optionally, the turbulent dynamic viscosity μ at the distal end is out It can be expressed as follows:

[0091] μ out =C out Vl

[0092] Where V is the combined velocity of the turbulence, l is the turbulence length scale, which is determined by the closest distance between the turbulence and the wall, and C out is the dynamic viscosity coefficient.

[0093] The dynamic viscosity coefficient can be expressed as follows:

[0094]

[0095] Among them, parameter C μ Parameter I G , parameter z G , parameter α is a preset value. In some embodiments, parameter C can be set μ =0.09, parameter I G =0.1, parameter z G =350, parameter α=0.22.

[0096] In an embodiment of the present application, the Li Chengling equation is embedded in the Reynolds time-averaged equation, so that the unknown quantity in the Reynolds time-averaged equation is converted into a turbulent viscosity variable, so that the closed Reynolds time-averaged equation can be easily and accurately solved to obtain the turbulent dynamic viscosity in the calculation domain. At the same time, the larger of the two turbulent dynamic viscosities corresponding to the near wall and the far end is used as the turbulent dynamic viscosity of the closed Reynolds time-averaged equation, so that the Reynolds time-averaged equation can more accurately describe the fluid motion law of the flow field in the calculation domain through the drag force of the stronger fluid, thereby more accurately reconstructing the flow field in the calculation domain and improving the accuracy of the flow field prediction.

[0097] In some embodiments, the loss function includes a Reynolds time-averaged residual term and a data residual term, and step A4 includes:

[0098] A41. Use the second coordinates corresponding to the above sampling points as input to the neural network model to be trained, and obtain the third flow field data corresponding to the above sampling points output by the above neural network model to be trained.

[0099] A42. Determine the Reynolds time-averaged residual corresponding to the sampling point according to the Reynolds time-averaged residual term in the loss function and the third flow field data.

[0100] A43. Determine the data residual corresponding to the sampling point according to the data residual term in the loss function, the second flow field data corresponding to the sampling point, and the third flow field data.

[0101] A44. Determine the loss value of the neural network model to be trained based on the Reynolds time-averaged residual and the data residual corresponding to the sampling points.

[0102] A45. Update the neural network model to be trained according to the loss value until the updated neural network model to be trained meets the preset requirements, thereby obtaining the trained neural network model.

[0103] Specifically, in the process of training the neural network model, the second coordinate corresponding to the sampling point is used as the input of the neural network model, and the air flow condition of the sampling point is predicted by the neural network model to obtain the third flow field data. Then, the Reynolds time-averaged residual of the sampling point is calculated by the Reynolds time-averaged residual term in the loss function and the predicted third flow field data, and the error between the predicted third flow field data and the measured second flow field data of the sampling point is calculated according to the data residual term in the loss function to obtain the data residual corresponding to the sampling point.

[0104] After calculating the Reynolds time-averaged residuals and data residuals of the sampling points, the loss value of the neural network model is determined comprehensively based on the Reynolds time-averaged residuals and data residuals, and the neural network model is updated according to the loss value until the updated neural network model meets the preset requirements (such as the number of iterations reaches the preset number of 5000 times), thereby obtaining the trained neural network model.

[0105] In some embodiments, the loss function of the neural network model may be in the following form:

[0106] L total =l f +l c

[0107] Among them, L total is the loss function of the neural network model, L f is the Reynolds time-averaged residual term, L c is the data residual term.

[0108] In some embodiments, the data residual term includes velocity residuals and pressure residuals in different spatial directions. The data residual term may be in the following form:

[0109] L c =L u +L v +L w +L p

[0110] L u , L v and L w Corresponding to the flow velocity residuals in the x, y and z directions, L p is the pressure residual. Correspondingly, the above loss function can be in the following form:

[0111] L total =L f +L u +L v +L w +L p

[0112] Optionally, the above-mentioned Reynolds time-averaged residual term can be in the following form:

[0113]

[0114] Where n is the index of the momentum residual function, f n (n can take values ​​of 1, 2, 3, or 4) corresponding to the momentum residual functions and continuity equation residual functions in the above-mentioned multiple spatial directions, such as f1, which is the momentum residual function in the x-direction, and f4, which is the continuity equation residual function combining multiple spatial directions;

[0115] i is the serial number of the sampling point corresponding to the training data, N is the number of sampling points corresponding to the training data, f i That is, the Reynolds time-averaged residual corresponding to the i-th sampling point, That is, the momentum residual in each spatial direction corresponding to the Reynolds time-averaged residual of the i-th sampling point.

[0116] From the above Reynolds time-averaged residual term, it can be seen that the Reynolds time-averaged residual term in the loss function contains the momentum residuals of each sampling point, and the Reynolds time-averaged residual of the sampling point is determined according to the square mean of each momentum residual.

[0117] Optionally, the pressure residual in the above data residual term can be in the following form:

[0118]

[0119] Among them, m is the label of the pressure residual function, x m (n values ​​include 1, 2, and 3) corresponding to the pressure residual function, where each pressure residual function can be in the following form:

[0120] x1=p zpb

[0121] Among them, x1 is the export boundary residual function, p zpb is the pressure at the outlet boundary of the predicted sampling point.

[0122]

[0123] Among them, x2 is the vertical symmetric boundary pressure residual function, p zpb is the predicted pressure at the vertical symmetry boundary.

[0124]

[0125] Among them, x3 is the horizontal symmetric boundary pressure residual function, p zpb is the predicted pressure at the horizontal symmetry boundary.

[0126] It can be understood that the above-mentioned pressure residual functions are constructed according to the boundary conditions of the calculation domain.

[0127] Optionally, in the data residual term, the velocity residual corresponding to the x-direction can be in the following form:

[0128]

[0129] Among them, l is the label of the preset residual function, y l (l values ​​include 1, 2, 3, 4, 5) corresponding to the preset residual functions, i is the serial number of the sampling point corresponding to the training data, N is the number of sampling points corresponding to the training data, y i That is, the data residual corresponding to the i-th sampling point, That is, the residuals corresponding to the data residuals of the i-th sampling point.

[0130] Optionally, the residual function y l The corresponding residual functions can be in the following forms:

[0131] y1=u isb -u initial

[0132]

[0133]

[0134] y4=u wb

[0135] y5=uu label

[0136] Among them, y1 is the residual function of the entrance boundary in the x direction of the computational domain, u isb is the predicted x-direction flow velocity at the inlet boundary at the sampling point, u initial is the known x-direction flow velocity at the inlet boundary of the computational domain;

[0137] y2 is the vertical symmetric boundary residual function in the x direction of the computational domain, u swb1 is the predicted flow velocity in the x direction of the vertical symmetric boundary at the sampling point;

[0138] y3 is the horizontal symmetric boundary residual function of the computational domain in the x direction, u swb2 is the predicted flow velocity in the x direction of the horizontal symmetric boundary at the sampling point;

[0139] y4 is the wall boundary residual function in the x direction of the computational domain, u wb is the predicted flow velocity in the x direction of the wall boundary at the sampling point;

[0140] y5 is the flow velocity residual function in the x direction, which is used to calculate the difference between the predicted third sampling data and the measured second sampling data in the x direction. u is the flow velocity in the x direction contained in the third flow field data corresponding to the sampling point. u label is the flow velocity in the x direction contained in the second flow field data corresponding to the sampling point.

[0141] It is understood that the flow velocity residual in the x-direction at a sampling point is determined based on the residuals of the various boundary conditions corresponding to that sampling point in the x-direction, as well as the residuals of the predicted third flow field data and the measured second flow field data. This fully considers the impact of the boundary conditions of the computational domain on the flow field data of the sampling point in the computational domain, thereby ensuring the accuracy of the flow field data predicted by the neural network model. Similarly, the flow velocity residuals in the y- and z-directions are also determined by the residual functions of the boundary conditions and the flow velocity residual functions corresponding to the y- and z-directions, and this will not be further described in detail in the present embodiment.

[0142] In some embodiments, the momentum residual corresponding to each spatial direction can be calculated based on different sampling points, and / or the data residual can be calculated based on different sampling points (for example, the pressure residual corresponding to each boundary can be calculated based on different sampling points, and / or the flow velocity residual corresponding to each spatial direction can be calculated based on different sampling points). That is, when calculating the momentum residual and / or data residual in each spatial direction, the calculation can be performed based on different sampling points. For example, when calculating the momentum residual in the x, y, and z directions, the momentum residual in the x direction of the sampling point can be calculated based on sampling points 1-100, the momentum residual in the y direction of the sampling point can be calculated based on sampling points 101-200, and the momentum residual in the z direction of the sampling point can be calculated based on sampling points 201-300. This allows the residuals in each direction (momentum residual and / or data residual) to be calculated based on different sampling points, eliminating the need to calculate each residual based on each sampling point, thereby reducing the amount of computation required during training.

[0143] In an embodiment of the present application, the Reynolds time-averaged residual and the data residual of the predicted third flow field data relative to the measured second flow field data are calculated based on the predicted value corresponding to the sampling point (i.e., the third flow field data). That is, the error of the predicted third flow field data is analyzed based on both physical laws and measured data, and then the neural network model is updated based on the more accurate loss value obtained comprehensively. This can reduce the difficulty of training the neural network model and improve the accuracy of the trained neural network model.

[0144] In some embodiments, the second flow field data and the third flow field data both include pressure and flow velocities in different spatial directions, where the flow velocities are used to reflect the air flow conditions at the sampling points in the corresponding spatial directions. Step A43 includes:

[0145] The residual of the pressure at the sampling point and the residual of the flow velocity in each of the spatial directions are calculated respectively based on the data residual term in the loss function, the second flow field data corresponding to the sampling point, and the third flow field data.

[0146] The data residual corresponding to the sampling point is determined according to the residual of the pressure at the sampling point, the residual of the flow velocity corresponding to each of the spatial directions, and the weight coefficient corresponding to each of the residuals.

[0147] Optionally, the above-mentioned multiple spatial directions may include the directions of the x, y and z coordinate axes in the earth space coordinate system, that is, the x direction, the y direction and the z direction.

[0148] Specifically, since the degree of influence of the flow velocity of the fluid in different spatial directions on the air flow condition is usually different, therefore, in the embodiment of the present application, in order to further improve the accuracy of the neural network model, the residual of the flow velocity in each spatial direction of the sampling point (i.e., the flow velocity residual) and the residual of the pressure of the sampling point (i.e., the pressure residual) are calculated respectively according to the data residual term in the loss function, the second flow field data corresponding to the sampling point, and the third flow field data. Then, the obtained residuals (i.e., the pressure residual and the flow velocity residual in the x-direction, the flow velocity residual in the y-direction, and the flow velocity residual in the z-direction) are respectively calculated with their corresponding weight coefficients to calculate the weighted residuals corresponding to each residual, and then the data residual corresponding to the sampling point is determined based on each weighted residual. Optionally, the weight coefficients corresponding to the above-mentioned respective residuals can be different weight coefficients to fully consider the degree of influence of the flow velocity in different spatial directions on the air flow condition and improve the prediction accuracy of the neural network model.

[0149] In some embodiments, the loss function of the neural network model may be in the following form:

[0150] L total =L f +w u L u +w v L v +w w L w +w p L p

[0151] Among them, L total is the loss function of the neural network model, L f is the Reynolds time-averaged residual term, L u , L v and L w Corresponding to the flow velocity residuals in the x-direction, y-direction and z-direction in the data residual term, w u 、w v 、w wThe weight coefficients corresponding to the flow velocity residuals in the x, y, and z directions, L p is the pressure residual in the data residual term, w p is the weight coefficient of the pressure residual.

[0152] Optionally, the weight coefficient w of the pressure residual p It can be set to 1. At this time, the above loss function can be in the following form:

[0153] L total =L f +w u L u +w v L v +w w L w +L p

[0154] In some embodiments, the weight coefficients corresponding to the data residual items in each of the above-mentioned spatial directions can be dynamic weights. During the training process of the neural network model, the weight coefficients corresponding to the data residual items in each spatial direction can be updated according to the relative error between the third flow field data of the sampling point predicted by the neural network model and the corresponding second flow field data. The dynamic weights are used to keep the weight coefficients corresponding to the data residual items in each spatial direction at an appropriate level to balance the various residual items, so that the trained neural network model can better reconstruct the flow field in the calculation domain and improve the accuracy of flow field prediction.

[0155] Optionally, the relative error between the third flow field data and the corresponding second flow field data can be expressed as follows:

[0156]

[0157] Among them, k i is the relative error of the flow velocity in the spatial direction between the third flow field data and the corresponding second flow field data corresponding to the sampling point, such as the relative error of the flow velocity u in the x direction, ‖·‖2 represents the second norm, Represents the vector composed of the second flow field data corresponding to the sampling point, Represents the vector composed of the third flow field data corresponding to the sampling point.

[0158] Optionally, when updating the weight coefficient of the data residual term in the corresponding spatial direction according to the relative error, the update can be performed according to the following formula:

[0159]

[0160] Among them, γ is a custom coefficient, i and j are the flow velocities corresponding to the spatial directions, which can be the flow velocity u in the x direction, the flow velocity v in the y direction, and the flow velocity w in the z direction. j (k j ) That is, take the smallest relative error among the relative errors of flow velocity corresponding to the spatial direction.

[0161] In an embodiment of the present application, when calculating the data residual corresponding to the sampling point, the final data residual of the sampling point is determined based on the residual of the flow velocity in each spatial direction and its corresponding weight coefficient, fully considering the differences in the degree of influence of the flow of fluid in different spatial directions on the air flow conditions, balancing the residual of the flow velocity in each spatial direction, thereby improving the accuracy of the neural network model prediction.

[0162] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0163] Example 2:

[0164] Corresponding to the flow field prediction method described in the above embodiment, Figure 3 A structural block diagram of a flow field prediction device provided in an embodiment of the present application is shown. For ease of explanation, only the parts related to the embodiment of the present application are shown.

[0165] Reference Figure 3 The device includes: a first coordinate acquisition module 31 and a prediction module 32.

[0166] The first coordinate acquisition module is used to obtain the first coordinate corresponding to the point to be measured, where the point to be measured is a coordinate point in the calculation domain.

[0167] A prediction module, configured to use the first coordinate as an input of a pre-trained neural network model to obtain first flow field data output by the neural network model, wherein the first flow field data reflects the air flow condition corresponding to the point to be measured;

[0168] The neural network model reconstructs the flow field in the calculation domain based on the physical laws described by the fluid control equations, predicts the air flow condition of the test point according to the flow field and the first coordinates, and obtains the first flow field data.

[0169] In the embodiments of the present application, since the fluid governing equations are embedded in the neural network model, the pre-trained neural network model can solve the embedded fluid governing equations and reconstruct the flow field within the computational domain based on the physical laws described by the fluid governing equations. Compared to solving the fluid governing equations through CFD simulation, this method does not require the construction of a large number of grids and is therefore not affected by grid resolution. Therefore, the pre-trained neural network model embedded with the fluid governing equations can quickly and accurately reconstruct the flow field within the computational domain and query the air flow conditions at the test point based on the reconstructed flow field and the first coordinate to obtain the required first flow field data. Furthermore, since the physical information constraints are embedded in the neural network model, the pre-trained neural network model can be obtained based on a small amount of measured data, compared to a purely data-driven neural network model. Furthermore, compared to fluid governing equations that do not consider the effects of temperature, humidity, and other factors in the actual environment, the neural network model embedded with the fluid governing equations is trained based on measured data and can reconstruct and predict the flow field in the computational domain based on both the physical laws described by the fluid governing equations and the actual conditions of the computational domain. This can improve the accuracy of flow field prediction and is beneficial for practical engineering applications.

[0170] In some embodiments, the flow field prediction device further includes:

[0171] The Reynolds residual function acquisition module is used to determine the Reynolds time-averaged residual function corresponding to the above-mentioned calculation domain according to the above-mentioned Reynolds time-averaged equation.

[0172] The loss function acquisition module is used to determine the loss function of the neural network model to be trained based on the Reynolds time-averaged residual function to obtain the neural network model to be trained.

[0173] The training data acquisition module is used to acquire training data, where the training data includes a plurality of second flow field data collected from sampling points in the calculation domain.

[0174] The training module is used to train the neural network model to be trained according to the training data until the trained neural network model meets the preset requirements, thereby obtaining the pre-trained neural network model.

[0175] In some embodiments, the Reynolds residual function acquisition module includes:

[0176] The momentum residual function determination unit is used to determine the momentum residual functions corresponding to multiple spatial directions according to the above-mentioned Reynolds time-averaged equation.

[0177] The Reynolds time-averaged residual function determining unit is configured to determine the Reynolds time-averaged residual function according to the momentum residual function corresponding to each of the spatial directions.

[0178] In some embodiments, the flow field prediction device further includes:

[0179] The Reynolds time-averaged equation determination module is used to determine the closed Reynolds time-averaged equation based on the Reynolds time-averaged equation and the zero equation.

[0180] Correspondingly, the momentum residual function determining unit is further configured to determine the momentum residual function corresponding to each of the spatial directions according to the closed Reynolds time-averaged equation.

[0181] In some embodiments, the Reynolds time-averaged equation determination module includes:

[0182] The Reynolds time-averaged equation closed unit is used to embed the Li Cheng zero equation into the above Reynolds time-averaged equation to obtain the closed Reynolds time-averaged equation, wherein the turbulent dynamic viscosity in the closed Reynolds time-averaged equation is determined according to the larger value of the turbulent dynamic viscosity near the wall and the turbulent dynamic viscosity at the far end in the above calculation domain.

[0183] In some embodiments, the loss function includes a Reynolds time-averaged residual term and a data residual term, and the training module includes:

[0184] The third flow field data acquisition unit is used to use the second coordinate corresponding to the above sampling point as the input of the above neural network model to be trained, and obtain the third flow field data corresponding to the above sampling point output by the above neural network model to be trained.

[0185] The Reynolds time-averaged residual determining unit is used to determine the Reynolds time-averaged residual corresponding to the sampling point according to the Reynolds time-averaged residual term in the loss function and the third flow field data.

[0186] A data residual determination unit is used to determine the data residual corresponding to the above sampling point based on the above data residual term in the above loss function, the second flow field data corresponding to the above sampling point, and the above third flow field data.

[0187] A loss value determination unit is used to determine the loss value of the neural network model to be trained based on the Reynolds time-averaged residual and the data residual corresponding to the sampling point.

[0188] An updating unit is used to update the above-mentioned neural network model to be trained according to the above-mentioned loss value until the updated above-mentioned neural network model to be trained meets the preset requirements, thereby obtaining the trained above-mentioned neural network model.

[0189] In some embodiments, the second flow field data and the third flow field data both include pressure and flow velocity in different spatial directions, and the flow velocity is used to reflect the air flow condition of the sampling point in the corresponding spatial direction. The training module includes:

[0190] The flow rate residual determination unit is used to calculate the residual of the pressure of the above-mentioned sampling point and the residual of the flow rate in each of the above-mentioned spatial directions according to the above-mentioned data residual term in the above-mentioned loss function, the second flow field data corresponding to the above-mentioned sampling point, and the above-mentioned third flow field data.

[0191] A weighting unit is used to determine the data residual corresponding to the sampling point according to the residual of the pressure at the sampling point, the residual of the flow velocity corresponding to each of the spatial directions, and the weight coefficient corresponding to each of the residuals.

[0192] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0193] Example 3:

[0194] Figure 4 This is a schematic diagram of the structure of a terminal device provided in one embodiment of the present application. Figure 4 As shown, the terminal device 4 of this embodiment includes: at least one processor 40 ( Figure 4 Only one processor is shown in the figure), a memory 41, and a computer program 42 stored in the memory 41 and executable on the at least one processor 40, wherein the processor 40 implements the steps of any of the above-mentioned method embodiments when executing the computer program 42.

[0195] The terminal device 4 can be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The terminal device may include, but is not limited to, a processor 40 and a memory 41. Those skilled in the art will understand that Figure 4 It is only an example of the terminal device 4 and does not constitute a limitation on the terminal device 4. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, it may also include input and output devices, network access devices, etc.

[0196] The processor 40 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. A general-purpose processor may be a microprocessor or any conventional processor.

[0197] In some embodiments, the memory 41 may be an internal storage unit of the terminal device 4, such as a hard disk or memory of the terminal device 4. In other embodiments, the memory 41 may also be an external storage device of the terminal device 4, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal device 4. Furthermore, the memory 41 may also include both an internal storage unit and an external storage device of the terminal device 4. The memory 41 is used to store an operating system, an application program, a boot loader, data, and other programs, such as the program code of the computer program. The memory 41 may also be used to temporarily store data that has been output or is to be output.

[0198] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0199] An embodiment of the present application also provides a network device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor implements the steps of any of the above-mentioned method embodiments when executing the computer program.

[0200] An embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the above-mentioned various method embodiments can be implemented.

[0201] An embodiment of the present application provides a computer program product. When the computer program product is run on a terminal device, the terminal device can implement the steps in the above-mentioned method embodiments when executing the computer program product.

[0202] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application implements all or part of the process of the above-mentioned method embodiment by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can at least include: any entity or device capable of carrying computer program code to the camera / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal, and software distribution medium. For example, a USB flash drive, mobile hard drive, magnetic disk, or optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals or telecommunication signals.

[0203] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0204] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0205] In the embodiments provided in this application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0206] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0207] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A flow field prediction method, characterized in that: include: Obtaining a first coordinate corresponding to a point to be measured, where the point to be measured is a coordinate point in the calculation domain; Using the first coordinate as an input of a pre-trained neural network model to obtain first flow field data output by the neural network model, wherein the first flow field data reflects the air flow condition corresponding to the point to be measured; The neural network model reconstructs the flow field in the calculation domain based on the physical laws described by the fluid control equation, predicts the air flow condition of the measured point according to the flow field and the first coordinates, and obtains the first flow field data; The fluid control equation is the Reynolds time-averaged equation, and the pre-trained neural network model is obtained according to the following steps: Determine the Reynolds time-averaged residual function corresponding to the calculation domain according to the Reynolds time-averaged equation; Determining a loss function of a neural network model to be trained based on the Reynolds time-averaged residual function to obtain the neural network model to be trained, wherein the loss function includes a Reynolds time-averaged residual term and a data residual term, and the Reynolds time-averaged residual term includes momentum residual functions corresponding to multiple spatial directions; Acquiring training data, wherein the training data includes a plurality of second flow field data collected from sampling points in the computational domain; Training the neural network model to be trained according to the training data, calculating a Reynolds time-averaged residual based on the Reynolds time-averaged residual term and the predicted value of the sampling point, and calculating a data residual based on the data residual term, the predicted value of the sampling point, and the measured value, wherein momentum residuals corresponding to different spatial directions in the Reynolds time-averaged residual are calculated based on different sampling points; The loss value of the neural network model to be trained is determined according to the Reynolds time-averaged residual and the data residual, and the neural network model to be trained is updated according to the loss value until the trained neural network model meets the preset requirements, thereby obtaining the pre-trained neural network model.

2. The flow field prediction method according to claim 1, wherein: Determining the Reynolds time-averaged residual function corresponding to the calculation domain according to the Reynolds time-averaged equation includes: Determine the momentum residual functions corresponding to multiple spatial directions respectively according to the Reynolds time-averaged equation; The Reynolds time-averaged residual function is determined according to the momentum residual function corresponding to each of the spatial directions.

3. The flow field prediction method according to claim 2, wherein: Before respectively determining the momentum residual functions corresponding to a plurality of spatial directions according to the Reynolds time-averaged equation, the method further includes: Determine the closed Reynolds time-averaged equation according to the Reynolds time-averaged equation and the zero equation; Correspondingly, determining the momentum residual functions corresponding to multiple spatial directions respectively according to the Reynolds time-averaged equation includes: The momentum residual function corresponding to each of the spatial directions is determined respectively according to the closed Reynolds time-averaged equation.

4. The flow field prediction method according to claim 3, wherein: The method of determining the closed Reynolds time-averaged equation based on the Reynolds time-averaged equation and the zero equation includes: The Li Chengling equation is embedded in the Reynolds time-averaged equation to obtain the closed Reynolds time-averaged equation, wherein the turbulent dynamic viscosity in the closed Reynolds time-averaged equation is determined according to the larger value of the turbulent dynamic viscosity near the wall and the turbulent dynamic viscosity at the far end in the calculation domain.

5. The flow field prediction method according to any one of claims 1 to 4, characterized in that: The training of the neural network model to be trained according to the training data, calculating the Reynolds time-averaged residual of the sampling point based on the Reynolds time-averaged residual term and the predicted value of the sampling point, and calculating the data residual based on the data residual term, the predicted value of the sampling point, and the measured value, includes: Using the second coordinate corresponding to the sampling point as the input of the neural network model to be trained, and obtaining third flow field data corresponding to the sampling point output by the neural network model to be trained; Calculating the momentum residual corresponding to each spatial direction according to the Reynolds time-averaged residual term in the loss function and the third flow field data corresponding to different sampling points, to obtain the Reynolds time-averaged residual; The data residual corresponding to the sampling point is determined according to the data residual term in the loss function, the second flow field data corresponding to the sampling point, and the third flow field data.

6. The flow field prediction method according to claim 5, characterized in that: The second flow field data and the third flow field data both include pressure and flow velocity in different spatial directions, the flow velocity being used to reflect the air flow condition of the sampling point in the corresponding spatial direction, and determining the data residual corresponding to the sampling point based on the data residual term in the loss function, the second flow field data corresponding to the sampling point, and the third flow field data, including: Calculating the residual of the pressure at the sampling point and the residual of the flow velocity in each of the spatial directions respectively according to the data residual term in the loss function, the second flow field data corresponding to the sampling point, and the third flow field data; The data residual corresponding to the sampling point is determined according to the residual of the pressure at the sampling point, the residual of the flow velocity corresponding to each of the spatial directions, and the weight coefficient corresponding to each of the residuals.

7. A flow field prediction device, characterized in that: include: A first coordinate acquisition module is used to acquire a first coordinate corresponding to a point to be measured, where the point to be measured is a coordinate point in the calculation domain; a prediction module, configured to use the first coordinate as an input of a pre-trained neural network model to obtain first flow field data output by the neural network model, wherein the first flow field data reflects the air flow condition corresponding to the point to be measured; The neural network model reconstructs the flow field in the calculation domain based on the physical laws described by the fluid control equation, predicts the air flow condition of the measured point according to the flow field and the first coordinates, and obtains the first flow field data; A Reynolds residual function acquisition module, configured to determine the Reynolds time-averaged residual function corresponding to the calculation domain according to the Reynolds time-averaged equation; a loss function determination module, configured to determine a loss function of a neural network model to be trained based on the Reynolds time-averaged residual function, to obtain the neural network model to be trained, wherein the loss function includes a Reynolds time-averaged residual term and a data residual term, and the Reynolds time-averaged residual term includes momentum residual functions corresponding to multiple spatial directions; A training data acquisition module, configured to acquire training data, wherein the training data includes a plurality of second flow field data collected from sampling points in the computational domain; a training module, configured to train the neural network model to be trained based on the training data, calculate a Reynolds time-averaged residual based on the Reynolds time-averaged residual term and the predicted value of the sampling point, and calculate a data residual based on the data residual term, the predicted value of the sampling point, and the measured value, wherein the momentum residuals corresponding to different spatial directions in the Reynolds time-averaged residual are calculated based on different sampling points; An updating module is used to determine the loss value of the neural network model to be trained based on the Reynolds time-averaged residual and the data residual, and to update the neural network model to be trained based on the loss value until the trained neural network model meets the preset requirements, thereby obtaining the pre-trained neural network model.

8. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images