Building wind environment training method and prediction method based on physical data hybrid drive

Abstract

The present invention relates to the technical field of building wind environment prediction, and specifically to a building wind environment training method and prediction method based on physical data hybrid drive. The total loss function of the present invention includes a data loss function and a physical loss function of a neural network model. Among them, the data loss function represents the difference between the building wind environment data predicted by the model and the calculated building wind environment data, and the physical loss function represents the difference in the mass change rate of the wind environment data predicted by the model and the wind environment calculated value in spatial position. The present invention adds a physical loss function on the basis of the data loss function to characterize the degree to which the neural network prediction value conforms to the physical laws, thereby improving the accuracy of the neural network prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of building wind environment prediction, and in particular to a building wind environment training method and a prediction method based on physical data hybrid drive. Background Art

[0002] After the known wind speed enters the environment where the building is located, the new wind speed formed at each position after the wind speed is affected by the environment. The new wind speed formed at each position in the environment is the building wind environment. Before or after the completion of the building, the building wind environment will be predicted based on the distribution location and building height of the building. The prior art uses a neural network model to predict the building wind environment, and before the model is used to predict the building wind environment, the model needs to be trained. The trained model can be used to predict the building wind environment. Because the loss function used in the prior art when training the model is the difference between the predicted value and the true value of the model, the prior art training model requires a large amount of training data, which increases the time required for training the model and reduces the model training speed.

[0003] In summary, the training method for the building wind environment prediction model in the prior art reduces the training speed.

[0004] Therefore, the prior art still needs to be improved and enhanced. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a building wind environment training method and a prediction method based on physical data hybrid drive, which solves the problem that the training method for the building wind environment prediction model in the prior art reduces the training speed.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] In a first aspect, the present invention provides a building wind environment training method based on physical data hybrid drive, which includes:

[0008] Constructing a training data set, wherein the training data set includes building form sample data and real building wind environment data corresponding to the building form sample data;

[0009] Applying a neural network model to the building form sample data to obtain building wind environment training data output by the neural network model;

[0010] Determine the loss sub-function of the neural network model based on the real data of the building wind environment and the training data of the building wind environment; determine the real change rate of the real data of the building wind environment; determine the training change rate of the training data of the building wind environment;

[0011] According to the true change rate, the training change rate and the loss sub-function, the total loss function of the neural network model is obtained. According to the total loss function, the neural network model is iteratively trained until the neural network model converges to obtain a building wind environment prediction model.

[0012] In one implementation, constructing a training data set includes:

[0013] Acquire building form three-dimensional data and real building wind environment data corresponding to the building form three-dimensional data;

[0014] The building form three-dimensional data is encoded to obtain a building form sample matrix, wherein the elements in the building form sample matrix are used to represent the spatial position, and the values ​​of the elements are used to represent the attribute information at the spatial position, wherein the attribute information includes building attribute information and non-building attribute, and the building form sample matrix is ​​used as building form sample data, and a training data set is constructed by the real building wind environment data and the building form sample data.

[0015] In one implementation, applying a neural network model to the building form sample data to obtain building wind environment training data output by the neural network model includes:

[0016] A neural network model is applied to the building form sample matrix to obtain adjacent point wind environment training data at the location of each adjacent point output by the neural network model, wherein each adjacent point is a point adjacent to a measuring point, the measuring point is a location where a non-building body is located, and the location of the measuring point corresponds to the center position of the building form sample matrix, and the adjacent point wind environment training data is used as the building wind environment training data.

[0017] In one implementation, determining the training change rate of the building wind environment training data includes:

[0018] Determine adjacent point wind environment training data for each adjacent point in the building wind environment training data;

[0019] Obtaining the spatial distance between the measuring point and each of the adjacent points;

[0020] The training change rate of the building wind environment training data is determined according to each of the spatial distances and the adjacent point wind environment training data of each of the adjacent points.

[0021] In one implementation, determining the training change rate of the building wind environment training data based on each of the spatial distances and the adjacent point wind environment training data of each of the adjacent points includes:

[0022] Determine each lateral distance, each longitudinal distance and each vertical distance in each of the spatial distances, and each of the lateral distances is equal, each of the longitudinal distances is equal, and each of the vertical distances is equal;

[0023] Determine the horizontal neighboring points, the vertical neighboring points and the vertical neighboring points among the adjacent points; determine the horizontal wind environment training data of the horizontal neighboring points, the vertical wind environment training data of the vertical neighboring points and the vertical wind environment training data of the vertical neighboring points among the adjacent point wind environment training data;

[0024] The training change rate of the building wind environment training data is determined based on the horizontal distance, the longitudinal distance, the vertical distance, the horizontal wind environment training data of the horizontal neighboring points, the vertical wind environment training data of the vertical neighboring points, and the vertical wind environment training data of the vertical neighboring points.

[0025] In one implementation, the measuring point is located at the center of an area enclosed by each of the adjacent points.

[0026] In one implementation, obtaining the total loss function of the neural network model based on the true change rate, the training change rate and the loss sub-function includes:

[0027] Subtract the true change rate from the training change rate to obtain the physical loss function of the neural network model;

[0028] The physical loss function and the loss sub-function are weighted to obtain a total loss function.

[0029] In a second aspect, an embodiment of the present invention further provides a prediction method, which includes:

[0030] Based on the above-mentioned physical data hybrid-driven building wind environment training method, a building wind environment prediction model is obtained;

[0031] Obtaining a building form matrix and incident wind data, wherein the elements in the building form matrix are used to represent the spatial position, and the values ​​of the elements are used to represent the building attribute information or non-building attribute located at the spatial position, and the incident wind is the wind incident to the building environment;

[0032] Applying a building wind environment prediction model to the building form matrix, obtains building wind environment prediction data.

[0033] In a third aspect, an embodiment of the present invention further provides a terminal device, wherein the terminal device includes a memory, a processor, and a building wind environment training program based on physical data hybrid drive stored in the memory and executable on the processor, and when the processor executes the building wind environment training program based on physical data hybrid drive, the steps of the building wind environment training method based on physical data hybrid drive are implemented;

[0034] Alternatively, the terminal device includes a memory, a processor, and a prediction program stored in the memory and executable on the processor, and when the processor executes the prediction program, the steps of the above-mentioned prediction method are implemented.

[0035] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, on which is stored a building wind environment training program based on physical data hybrid drive, and when the building wind environment training program based on physical data hybrid drive is executed by a processor, the steps of the building wind environment training method based on physical data hybrid drive are implemented;

[0036] Alternatively, a prediction program is stored on the computer-readable storage medium, and when the prediction program is executed by the processor, the steps of the above-mentioned prediction method are implemented.

[0037] Beneficial effect: When training the neural network model, the total loss function used in the present invention includes the loss sub-function, true change rate and training change rate of the neural network model. Among them, the loss sub-function represents the difference between the building wind environment data predicted by the model and the actual building wind environment data, the true change rate represents the change information of the actual building wind environment data with position, and the training change rate represents the change information of the building wind environment data predicted by the model with position. From the above analysis, it can be seen that the present invention adds the true change rate and the training change rate on the basis of the loss sub-function. The two can characterize the degree of physical loss of the neural network model. Since the degree of physical loss of the model is taken into account when calculating the total loss function, the accuracy of the calculated total loss function is improved. Therefore, the present invention uses less training data sets to train the model, while ensuring the accuracy of the model after training. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is the overall flow chart of the present invention;

[0039] Figure 2 It is a schematic diagram of the measuring points and adjacent points of the present invention;

[0040] Figure 3 A structural diagram of a building wind environment training device based on physical data hybrid drive provided by the present invention;

[0041] Figure 4 This is a block diagram of the internal structure principle of the terminal device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The following is a clear and complete description of the technical solution of the present invention in combination with the embodiments and the accompanying drawings. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] It has been found through research that after a known wind speed enters the environment where a building is located, the wind speed is affected by the environment and forms new wind speeds at various locations. The new wind speeds formed at various locations within the environment are the building wind environment. Before or after the completion of the building, the building wind environment will be predicted based on the distribution location and building height of the building. The prior art uses a neural network model to predict the building wind environment, and before the model is used to predict the building wind environment, the model needs to be trained. The trained model can be used to predict the building wind environment. The prior art requires a large amount of training data to train the model because the loss function used in the training model is the difference between the predicted value and the true value of the model, which increases the time required to train the model and reduces the model training speed.

[0044] In order to solve the above technical problems, the present invention provides a building wind environment training method and a prediction method based on physical data hybrid drive, which solves the problem that the training method for the building wind environment prediction model in the prior art reduces the training speed.

[0045] Embodiment 1: The building wind environment training method based on physical data hybrid drive of this embodiment can be applied to a terminal device, and the terminal device can be a terminal product with data processing function, such as a computer. Figure 1 As shown in , the building wind environment training method based on physical data hybrid drive specifically includes the following steps:

[0046] S100, constructing a training data set, wherein the training data set includes building form sample data and real building wind environment data corresponding to the building form sample data;

[0047] S200, applying a neural network model to the building form sample data to obtain building wind environment training data output by the neural network model;

[0048] S300, determining the loss sub-function of the neural network model according to the real data of the building wind environment and the training data of the building wind environment; determining the real change rate of the real data of the building wind environment; determining the training change rate of the training data of the building wind environment;

[0049] S400, obtaining the total loss function of the neural network model based on the true change rate, the training change rate and the loss sub-function, and iteratively training the neural network model based on the total loss function until the neural network model converges to obtain a building wind environment prediction model.

[0050] The real data of building wind environment is also the calculated data of building wind environment. The training change rate and the real change rate are collectively referred to as the mass change rate, which is the change rate of air mass inflow and outflow at any particle in space. In the calculation of incompressible fluid, the mass change rate of air is 0. The loss sub-function is the difference between the mass change rate of predicted data and the mass change rate of calculated data.

[0051] In this embodiment, step S100 includes the following specific steps S101 and S102:

[0052] S101, acquiring building form three-dimensional data and real building wind environment data corresponding to the building form three-dimensional data.

[0053] An urban building form case is selected from the urban information database of the geographic information system GIS. The selected range can be set to 300 meters to 600 meters as needed. In this embodiment, a square is used to represent the range, that is, the side length of the square can be set to 300 meters to 600 meters, and is set to 300 meters in this embodiment. The selected urban building form case includes the three-dimensional data of the building form at various locations in the city (including whether there are buildings and the building height) and the wind speed at different heights at various locations (that is, the real data of the building wind environment), wherein computational fluid dynamics CFD (CFD includes the software package Fluent) is applied to the urban building form case to obtain the real data of the building wind environment.

[0054] The 3D data of building form includes the various points in the city involved in the above-mentioned urban building form cases. 3D data of building form (with points representing selected locations) ,in, For the first point The horizontal axis of For point The vertical coordinate of For architecture, then is the building height. If point If there is no building, Zero. Each building form 3D data The actual data of the building wind environment is , Representatives in the city Altitude The wind speed at the place.

[0055] If the urban building form cases (the selected cases involve a square with a side length of 300 meters) are rasterized at a resolution of 3 meters, 100x100 grids can be divided, and then sampling is performed with each grid as the center and (241, 241) as the sampling frame to obtain 10,000 sets of sampling data. Each set of sampling data includes 241x241 building form three-dimensional data. 241x241 building form three-dimensional data constitute a set of building form three-dimensional data.

[0056] S102, encode the building form three-dimensional data to obtain a building form sample matrix, wherein the elements in the building form sample matrix are used to represent the spatial position, and the values ​​of the elements are used to represent the attribute information at the spatial position, wherein the attribute information includes building attribute information and non-building attribute, and the building form sample matrix is ​​used as building form sample data, and a training data set is constructed by the real building wind environment data and the building form sample data.

[0057] Each set of building form three-dimensional data is encoded to obtain a building form sample matrix X. If the value of the element in the 120th row and 126th column of the building form sample matrix X is 0, it means that there is no building at the spatial position corresponding to the 120th row and 126th column; if the value of the element in the 120th row and 126th column is 90, it means that there is a building at the spatial position corresponding to the 120th row and 126th column, and the height of the building is 90 meters.

[0058] In this embodiment, step S200 includes the following specific steps: applying a neural network model to the building form sample matrix to obtain adjacent point wind environment training data at the location of each adjacent point output by the neural network model, each adjacent point is a point adjacent to a measuring point, the measuring point is a location where a non-building body is located, the location of the measuring point corresponds to the center position of the building form sample matrix, and the adjacent point wind environment training data is used as the building wind environment training data.

[0059] The measurement point in this example corresponds to the spatial position corresponding to the center of the architectural form sample matrix X, that is, the measurement point and the center of X have the same horizontal and vertical coordinates. For example, the spatial position corresponding to the center of X is , then the position of the measuring point is ,in Generated by the neural network model, that is, the neural network model will automatically select the vertical coordinates to be output Building wind environment data at the location.

[0060] The adjacent points of the measuring point are Figure 2As shown, in this embodiment, the subsequent calculation of the training change rate (pre represents the wind speed change at adjacent points) Figure 2 The number of adjacent points in , but the number of adjacent points is not limited to six. Figure 2 As shown, the six adjacent points are , , , , , Among them, the measuring point Move one unit along the negative x-axis (one unit is 3 meters in this example) to get ; Measurement point Move one unit along the positive x-axis to get ; Measurement point Move one unit along the negative y-axis to get ; Measurement point Move one unit along the positive y-axis to get ; Measurement point Move one unit along the negative z-axis to get ; Measurement point Move one unit along the positive z-axis to get .

[0061] In this embodiment, the training change rate in step S300 The method comprises the following specific steps S301, S302, S303, and S304:

[0062] S301, determining adjacent point wind environment training data of each adjacent point in the building wind environment training data.

[0063] The adjacent point wind environment training data in this embodiment is the wind speed of each adjacent point output by the neural network model based on the building form sample matrix X.

[0064] S302, determining each lateral distance in each of the spatial distances , each longitudinal distance And each vertical distance , and each of the lateral distances is equal, each of the longitudinal distances is equal, and each of the vertical distances is equal.

[0065] Adjacent points With measuring point The distance along the x-axis is the horizontal distance , adjacent points With measuring point The distance along the x-axis is also the lateral distance . Adjacent points With measuring point The distance along the y-axis is the longitudinal distance , adjacent points With measuring point The distance along the y-axis is also the longitudinal distance . Adjacent points With measuring point The distance along the z axis is the vertical distance , adjacent points With measuring point The distance along the z-axis is also the vertical distance .

[0066] S303, determining the horizontal neighboring points, the longitudinal neighboring points and the vertical neighboring points among the adjacent points; determining the horizontal wind environment training data of the horizontal neighboring points, the longitudinal wind environment training data of the longitudinal neighboring points, and the vertical wind environment training data of the vertical neighboring points among the adjacent point wind environment training data.

[0067] Lateral neighbors include and , the vertical neighbors include and , vertical neighbors include and .

[0068] The wind environment training data of the lateral neighboring points in the lateral direction includes and ,in For lateral neighbors Training wind speed in the lateral direction, For lateral neighbors Training wind speed in the lateral direction.

[0069] The wind environment training data of the longitudinal neighboring points in the longitudinal direction includes and ,in Vertical neighbor Training wind speed in the longitudinal direction, Vertical neighbor Training wind speed in the longitudinal direction.

[0070] The vertical wind environment training data of the vertical neighboring points includes and ,in Vertical neighbor Training wind speed in the vertical direction, Vertical neighbor Training wind speed in the vertical direction.

[0071] All training wind speeds in this embodiment are wind speeds output by the neural network model.

[0072] S304, determining a training change rate of the building wind environment training data according to the horizontal distance, the vertical distance, the vertical distance, the horizontal wind environment training data of the horizontal neighboring point, the vertical wind environment training data of the vertical neighboring point, and the vertical wind environment training data of the vertical neighboring point. .

[0073] ;

[0074] In this embodiment, the actual change rate in step S300 :

[0075] ;

[0076] In the formula, For lateral neighbors The actual wind speed in the lateral direction (that is, the building form sample matrix X corresponding to True wind speed in the lateral direction), For lateral neighbors The actual wind speed in the lateral direction (that is, the building form sample matrix X corresponding to true wind speed in the lateral direction). Vertical neighbor The actual wind speed in the longitudinal direction (that is, the building form sample matrix X corresponding to True wind speed in the longitudinal direction), Vertical neighbor The actual wind speed in the longitudinal direction (that is, the building form sample matrix X corresponding to true wind speed in the longitudinal direction). Vertical neighbor The actual wind speed in the vertical direction (that is, the building form sample matrix X corresponding to True wind speed in the vertical direction), Vertical neighbor The actual wind speed in the vertical direction (that is, the building form sample matrix X corresponding to True wind speed in the vertical direction).

[0077] In this embodiment, the loss subfunction :

[0078] ;

[0079] n is the total number of sampling points in a cube formed by a measuring point and all its adjacent points (the sampling points include Figure 2 The measured points and adjacent points in the cube also include other points in the cube). For the The actual wind speed at each sampling point is The output of the neural network model The training wind speed at each sampling point.

[0080] The total loss function in this embodiment :

[0081] ;

[0082] In the formula, is a hyperparameter less than 1.

[0083] Embodiment 2, based on embodiment 1, provides a prediction method for predicting the wind environment of a building. The embodiment includes the following specific steps: obtaining a building form matrix and incident wind data, wherein the elements in the building form matrix are used to characterize the spatial position, and the values ​​of the elements are used to characterize the building attribute information or non-building attributes located at the spatial position, and the incident wind is the wind incident to the building environment; the building wind environment prediction model is applied to the building form matrix to obtain the building wind environment prediction data.

[0084] The incident wind data of this embodiment is 2 meters per second, that is, if the wind speed enters the environment where the building is located (for example, the environment is a park with several buildings) at a speed of 2 meters per second, the wind speed at each location point in the environment is predicted.

[0085] The first step is to process the calculation range according to the resolution of 3m grid. Taking each grid point as the center, sample according to the data sampling frame (241, 241) to form a set of building form coding matrix data to be calculated (that is, building form matrix ).

[0086] In the second step, each building form coding matrix is ​​input into the neural network model in turn to obtain the building wind speed matrix (3, 3, 3, 25) of the corresponding point. The first 3 is The second 3 is The third 3 represents the three wind directions, and 25 represents the 25 vertical sampling points. Points involving buildings will be skipped.

[0087] In the third step, after the process in step 2, each point to be calculated (i.e., measurement point) will be calculated up to 9 times. The multiple wind speed results calculated for each point are accumulated and averaged according to the number of calculations, and finally the wind speed distribution result within the range to be calculated is obtained.

[0088] In summary, the model input of the present invention does not have to be limited to only inputting spatial position information, but can be replaced by inputting wind field information (building space form). After the model structure changes, the model can learn and predict wind fields of different building forms. The model output of the present invention becomes to output the wind speed of multiple measuring points at the same time. For the entire wind field, each measuring point will be calculated multiple times and the average value will be taken, which can improve the accuracy of model prediction. The model of the present invention can also be used to calculate the building acoustic environment and building thermal environment.

[0089] This embodiment also provides a building wind environment training device based on physical data hybrid drive, such as Figure 3 As shown, the device comprises:

[0090] The data set construction module 01 is used to construct a training data set, wherein the training data set includes building form sample data and real building wind environment data corresponding to the building form sample data;

[0091] The training data generation module 02 is used to apply the neural network model to the building form sample data to obtain the building wind environment training data output by the neural network model;

[0092] The loss function and change rate calculation module 03 is used to determine the loss sub-function of the neural network model according to the real data of the building wind environment and the training data of the building wind environment; determine the real change rate of the real data of the building wind environment; determine the training change rate of the training data of the building wind environment;

[0093] The training module 04 is used to obtain the total loss function of the neural network model based on the true change rate, the training change rate and the loss sub-function, and iteratively train the neural network model based on the total loss function until the neural network model converges to obtain a building wind environment prediction model.

[0094] Based on the above embodiments, the present invention further provides a terminal device, whose principle block diagram can be shown as follows: Figure 4 As shown. The terminal device includes a processor, a memory, a network interface, and a display screen connected via a system bus. Among them, the processor of the terminal device is used to provide computing and control capabilities. The memory of the terminal device includes 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 the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the terminal device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a building wind environment training method based on physical data hybrid drive is implemented. The display screen of the terminal device can be a liquid crystal display screen or an electronic ink display screen.

[0095] Those skilled in the art will understand that Figure 4 The principle block diagram shown in the figure is only a block diagram of a partial structure related to the scheme of the present invention, and does not constitute a limitation on the terminal device to which the scheme of the present invention is applied. The specific terminal device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0096] In one embodiment, a terminal device is provided, the terminal device comprising a memory, a processor, and a building wind environment training method program based on physical data hybrid drive stored in the memory and executable on the processor, and when the processor executes the building wind environment training method program based on physical data hybrid drive, the following operation instructions are implemented:

[0097] Constructing a training data set, wherein the training data set includes building form sample data and real building wind environment data corresponding to the building form sample data;

[0098] Applying a neural network model to the building form sample data to obtain building wind environment training data output by the neural network model;

[0099] Determine the loss sub-function of the neural network model based on the real data of the building wind environment and the training data of the building wind environment; determine the real change rate of the real data of the building wind environment; determine the training change rate of the training data of the building wind environment;

[0100] According to the true change rate, the training change rate and the loss sub-function, the total loss function of the neural network model is obtained. According to the total loss function, the neural network model is iteratively trained until the neural network model converges to obtain a building wind environment prediction model.

[0101] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided by the present invention can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0102] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention 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. However, 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 embodiments of the present invention.

Claims

1. A building wind environment training method based on physical data hybrid drive, characterized in that: include: Constructing a training data set, wherein the training data set includes building form sample data and real building wind environment data corresponding to the building form sample data; Applying a neural network model to the building form sample data to obtain building wind environment training data output by the neural network model; Determine the loss sub-function of the neural network model based on the real data of the building wind environment and the training data of the building wind environment; determine the real change rate of the real data of the building wind environment; Determining a training change rate of the building wind environment training data; According to the true change rate, the training change rate and the loss sub-function, a total loss function of the neural network model is obtained, and according to the total loss function, the neural network model is iteratively trained until the neural network model converges to obtain a building wind environment prediction model; The applying a neural network model to the building form sample data to obtain building wind environment training data output by the neural network model includes: Applying a neural network model to a building form sample matrix to obtain adjacent point wind environment training data at the location of each adjacent point output by the neural network model, wherein the building form sample matrix is ​​an encoding of the building form three-dimensional data, each adjacent point is a point adjacent to a measuring point, the measuring point is a location where a non-building body is located, and the location of the measuring point corresponds to the center position of the building form sample matrix, and the adjacent point wind environment training data is used as the building wind environment training data; Determining the training change rate of the building wind environment training data includes: Determine adjacent point wind environment training data for each adjacent point in the building wind environment training data; Obtaining the spatial distance between the measuring point and each of the adjacent points; Determine each lateral distance, each longitudinal distance and each vertical distance in each of the spatial distances, and each of the lateral distances is equal, each of the longitudinal distances is equal, and each of the vertical distances is equal; Determine the horizontal neighboring points, the vertical neighboring points and the vertical neighboring points among the adjacent points; determine the horizontal wind environment training data of the horizontal neighboring points, the vertical wind environment training data of the vertical neighboring points and the vertical wind environment training data of the vertical neighboring points among the adjacent point wind environment training data; The training change rate of the building wind environment training data is determined based on the horizontal distance, the longitudinal distance, the vertical distance, the horizontal wind environment training data of the horizontal neighboring points, the vertical wind environment training data of the vertical neighboring points, and the vertical wind environment training data of the vertical neighboring points.

2. The building wind environment training method based on physical data hybrid drive according to claim 1, characterized in that: The constructing of the training data set comprises: Acquire building form three-dimensional data and real building wind environment data corresponding to the building form three-dimensional data; The building form three-dimensional data is encoded to obtain a building form sample matrix, wherein the elements in the building form sample matrix are used to represent the spatial position, and the values ​​of the elements are used to represent the attribute information at the spatial position, wherein the attribute information includes building attribute information and non-building attribute, and the building form sample matrix is ​​used as building form sample data, and a training data set is constructed by the real building wind environment data and the building form sample data.

3. The building wind environment training method based on physical data hybrid drive according to claim 1 or 2, characterized in that: The measuring point is located at the center of the area surrounded by each of the adjacent points.

4. The building wind environment training method based on physical data hybrid drive according to claim 1, characterized in that: The total loss function of the neural network model is obtained according to the true change rate, the training change rate and the loss sub-function, including: Subtract the true change rate from the training change rate to obtain the physical loss function of the neural network model; The physical loss function and the loss sub-function are weighted to obtain a total loss function.

Images