A wind environment rapid prediction method based on physical consistency neural network

Abstract

The present application relates to the technical field of building wind environment prediction, in particular to a wind environment fast prediction method based on physical consistency neural network. The present application uses a training data set containing building shape sample data and building wind environment real data to train a neural network model, obtains a building wind environment prediction model, and the building wind environment prediction model is the trained neural network model. Then, the shape matrix of the building whose wind environment needs to be predicted and incident wind data are obtained, the building wind environment prediction model is applied to the shape matrix, when the incident wind data enters the building wind environment, the wind speed of each position point in the environment is predicted to obtain building wind environment prediction data, and the wind environment prediction data is drawn into a wind environment prediction result graph, so that the prediction data can be conveniently and intuitively viewed. The present application can predict the wind speed of each position point in the building environment in advance, so as to give corresponding measures according to the predicted wind speed in advance.

Description

[0001] This invention is a divisional application of the application filed on September 12, 2024, with application number 2024112752402, entitled "A Training Method and Prediction Method for Building Wind Environment Based on Hybrid Physical Data," originally filed in China. Technical Field

[0002] This invention relates to the field of building wind environment prediction technology, specifically to a rapid wind environment prediction method based on a physical consistency neural network. Background Technology

[0003] When a known wind speed enters the environment surrounding a building, it is influenced by that environment, resulting in new wind speeds at various locations. These new wind speeds at different locations within the environment constitute the building's wind environment. Before or after building completion, the building's wind environment is predicted based on its location and height. Current technology uses neural network models to predict the building's wind environment. However, these models require training before they can be used for prediction. Because the loss function used in training the model is the difference between the model's predictions and the actual values, current technology requires a large amount of training data, increasing the training time and thus reducing the model's training speed.

[0004] In summary, existing training methods for building wind environment prediction models reduce training speed.

[0005] Therefore, existing technologies still need to be improved and enhanced. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides a rapid wind environment prediction method based on a physically consistent neural network, which solves the problem that the training methods for building wind environment prediction models in the prior art reduce the training speed.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention provides a method for rapid wind environment prediction based on a physically consistent neural network, comprising:

[0009] Construct a training dataset, which includes building form sample data and the corresponding real building wind environment data;

[0010] A neural network model is applied to the architectural form sample data to obtain architectural wind environment training data output by the neural network model.

[0011] Based on the real building wind environment data and the building wind environment training data, determine the loss function of the neural network model; determine the true rate of change of the real building wind environment data; determine the training rate of change of the building wind environment training data;

[0012] Based on the actual rate of change, the training rate of change, and the loss function, the total loss function of the neural network model is obtained. Based on the total loss function, the neural network model is iteratively trained until the neural network model converges, thus obtaining the building wind environment prediction model.

[0013] Obtain building form matrix and incident wind data. The elements in the building form matrix are used to represent spatial location, and the values ​​of the elements are used to represent building attribute information or non-building attributes located at the spatial location. Incident wind is the wind that is incident on the building environment.

[0014] The building wind environment prediction model is applied to the building form matrix to obtain building wind environment prediction data.

[0015] In one implementation, the training rate of change is Mass pre :

[0016] Masspre = (ux 0+1 -ux0-1)×(Δy×Δz)+(vy 0+1 -vy0-1)×(Δx×Δz)+(wz 0+1 -wz0-1)×(Δy×Δx)

[0017] In the formula, (x0, y0, z0) represents the spatial location of the measuring point, x0-1, x0+1, y0-1, y0+1, z0-1, and z0+1 are the locations of the six adjacent points, and u x0+1 Let u be the training wind speed of the lateral neighbor x0+1 in the lateral direction. x0-1 Let v be the training wind speed of the lateral neighbor x0-1 in the lateral direction. y0+1 Let v be the training wind speed of the longitudinal neighbor y0+1 in the longitudinal direction. y0-1 Let w be the training wind speed of the longitudinal neighbor point y0-1 in the longitudinal direction. z0+1 For the training wind speed of the vertical neighbor z0+1 in the vertical direction, w z0-1 Let z0-1 be the training wind speed in the vertical direction of the adjacent point z0-1, Δx be the lateral distance between the adjacent point z0-1 and the measuring point (x0,y0,z0) along the x-axis, Δy be the longitudinal distance between the adjacent point y0-1 and the measuring point (x0,y0,z0) along the y-axis, and Δz be the vertical distance between the adjacent point z0+1 and the measuring point (x0,y0,z0) along the z-axis.

[0018] In one implementation, the actual rate of change is Masscal :

[0019] Masscal=(u′x 0+1 -u′x0-1)×(Δy×Δz)+(v′y 0+1 -v′y0-1)×(Δx×Δz)+(w′z 0+1 -w′z0-1)×(Δy×Δx)

[0020] In the formula, u′ x0+1 Let u′ be the actual wind speed at the lateral neighbor x0+1 in the lateral direction. x0-1 Let v′ be the true wind speed in the lateral direction of the lateral neighbor x0-1. y0+1 Let v′ be the true wind speed at the longitudinal neighbor point y0+1 in the longitudinal direction. y0-1 Let w′ be the true wind speed at the longitudinal neighbor point y0-1 in the longitudinal direction. z0+1 Let w′ be the true wind speed at the vertical neighbor point z0+1 in the vertical direction. z0-1 Let z0-1 be the actual wind speed in the vertical direction of the vertical neighboring point z0-1.

[0021] In one implementation, the method further includes: the incident wind data is introduced into the building wind environment to predict the wind speed at various locations within the environment.

[0022] One implementation also includes:

[0023] Computational Fluid Dynamics (CFD) is applied to urban building morphology cases to obtain real-world wind environment data for the buildings. These urban building morphology cases include three-dimensional data of building morphology at various locations in the city and wind speeds at different heights at each location.

[0024] Beneficial Effects: In training the neural network model, this invention uses a total loss function that includes the model's loss sub-function, the true rate of change, and the training rate of change. The loss sub-function represents the difference between the model's predicted building wind environment data and the actual building wind environment data; the true rate of change represents the location-dependent changes in the actual building wind environment data; and the training rate of change represents the location-dependent changes in the model's predicted building wind environment data. As the above analysis shows, this invention adds the true rate of change and the training rate of change to the loss sub-function. These two factors characterize the physical loss of the neural network model. Because the physical loss of the model is considered when calculating the total loss function, the accuracy of the calculated total loss function is improved. Therefore, this invention uses a smaller training dataset to train the model while still ensuring the accuracy of the trained model. Attached Figure Description

[0025] Figure 1 This is an overall flowchart of the present invention;

[0026] Figure 2 This is a schematic diagram of the measuring points and adjacent points of the present invention;

[0027] Figure 3 This is a structural diagram of the building wind environment training device based on physical data hybrid driving provided by the present invention.

[0028] Figure 4 This is a block diagram illustrating the internal structure of a terminal device provided in an embodiment of the present invention. Detailed Implementation

[0029] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments and accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0030] Research has found that when a known wind speed enters the environment surrounding a building, it is influenced by that environment, resulting in new wind speeds at various locations. These new wind speeds at different locations within the environment constitute the building's wind environment. Before or after building completion, the building's wind environment is predicted based on its location and height. Current technology uses neural network models to predict the building's wind environment. However, these models require training before they can be used for predictions. Because the loss function used in training the model is the difference between the model's predictions and the actual values, current technology requires a large amount of training data, increasing the training time and thus reducing the model's training speed.

[0031] To address the aforementioned technical problems, this invention provides a rapid wind environment prediction method based on a physically consistent neural network, which solves the problem that the training methods for building wind environment prediction models in the prior art reduce the training speed.

[0032] Example 1 provides a training method for building wind environment based on physical data hybrid driving. This training method can be applied to a terminal device, which can be a terminal product with data processing capabilities, such as a computer. In this example, as... Figure 1 As shown, the building wind environment training method based on physical data hybrid driving specifically includes the following steps:

[0033] S100, Construct a training dataset, which includes building form sample data and the corresponding real building wind environment data;

[0034] S200, Apply a neural network model to the building form sample data to obtain the building wind environment training data output by the neural network model;

[0035] S300, based on the real building wind environment data and the building wind environment training data, determine the loss function of the neural network model; determine the real rate of change of the real building wind environment data; determine the training rate of change of the building wind environment training data;

[0036] S400: Based on the actual rate of change, the training rate of change, and the loss function, the total loss function of the neural network model is obtained. Based on the total loss function, the neural network model is iteratively trained until the neural network model converges, thereby obtaining the building wind environment prediction model.

[0037] Real-world building wind environment data refers to calculated building wind environment data. The training rate of change and the actual rate of change are collectively referred to as the mass rate of change, which is the rate of change of air mass flowing in and out of any point in space. In incompressible fluid calculations, the mass rate of change of air is 0. The loss function is the difference between the mass rate of change of the predicted data and the mass rate of change of the calculated data.

[0038] In this embodiment, step S100 includes the following specific steps S101 and S102:

[0039] S101, acquire three-dimensional data of the building form and the real data of the building wind environment corresponding to the three-dimensional data of the building form.

[0040] Urban building morphology cases are selected from the city information database of the Geographic Information System (GIS). The selection range can be set from 300 meters to 600 meters as needed. In this example, the range is represented by a square, meaning the side length of the square can be set from 300 meters to 600 meters; in this example, it is set to 300 meters. The selected urban building morphology cases include three-dimensional data of building morphology at various locations in the city (including whether there are buildings and their height) and wind speeds at different heights at each location (i.e., real-world wind environment data). Computational Fluid Dynamics (CFD) (including the software package Fluent) is applied to the urban building morphology cases to obtain real-world wind environment data.

[0041] The 3D data of building form includes the 3D data of the building form of each point i (representing the selected location) within the city involved in the above urban building form case studies (x). i ,y i ,h i ), where x i Let y be the x-coordinate of the i-th point. i Let h be the ordinate of point i. If point i is a building, then hi Let h be the height of the building. If there is no building at point i, then h... i Zero. Three-dimensional data for each building form (x...) i ,y i ,h i Each corresponds to a real data point of architectural wind environment, s. i,j s i,j This represents the wind speed at point i at altitude j within the city.

[0042] If the urban building form case (the selected case involves a square with a side length of 300 meters) is rasterized with a resolution of 3 meters, then it can be divided into 100x100 grids. Then, with each grid as the center and (241,241) as the sampling frame, sampling is performed to obtain 10,000 sets of sampling data. Each set of sampling data includes 241x241 three-dimensional building form data. The 241x241 three-dimensional building form data constitute a set of three-dimensional building form data.

[0043] S102, the three-dimensional data of the building form is encoded to obtain a building form sample matrix. The elements in the building form sample matrix are used to represent spatial location, and the values ​​of the elements are used to represent attribute information at the spatial location. The attribute information includes building attribute information and non-building attributes. The building form sample matrix is ​​used as building form sample data, and a training dataset is constructed from the real building wind environment data and the building form sample data.

[0044] Each set of 3D building form 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 location 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 location corresponding to the 120th row and 126th column, and the height of the building is 90 meters.

[0045] In this embodiment, step S200 includes the following specific steps: applying a neural network model to the building form sample matrix to obtain the 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 not where the building is located, the position 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 building wind environment training data.

[0046] In this example, the measurement point corresponds to the spatial position of the center of the building form sample matrix X. That is, the measurement point and the center of X have the same horizontal and vertical coordinates. For example, if the spatial position corresponding to the center of X is (x0, y0), then the position of the measurement point is (x0, y0, z0). Here, z0 is generated by the neural network model. In other words, the neural network model will automatically select the building wind environment data at the vertical coordinate z0 that needs to be output.

[0047] Adjacent points of the measuring point, such as Figure 2 As shown, in this embodiment, the subsequent calculation of the training rate of change Mass pre (pre represents the wind speed change between adjacent points) used Figure 2 The six adjacent points in the array, but the number of adjacent points is not limited to six. For example... Figure 2 As shown, the six adjacent points are x0-1, x0+1, y0-1, y0+1, z0-1, and z0+1. Specifically, the measuring point (x0, y0, z0) is moved one unit along the negative x-axis (in this embodiment, one unit is 3 meters) to obtain x0-1; the measuring point (x0, y0, z0) is moved one unit along the positive x-axis to obtain x0+1; the measuring point (x0, y0, z0) is moved one unit along the negative y-axis to obtain y0-1; the measuring point (x0, y0, z0) is moved one unit along the positive y-axis to obtain y0+1; the measuring point (x0, y0, z0) is moved one unit along the negative z-axis to obtain z0-1; and the measuring point (x0, y0, z0) is moved one unit along the positive z-axis to obtain z0+1.

[0048] In this embodiment, the training rate of change Mass in step S300 pre The specific steps are as follows: S301, S302, S303, and S304.

[0049] S301, determine the adjacent point wind environment training data of each of the adjacent points in the building wind environment training data.

[0050] In this embodiment, the wind environment training data for adjacent points is the wind speed of each adjacent point output by the neural network model based on the building form sample matrix X.

[0051] S302, determine each horizontal distance Δx, each vertical distance Δy, and each vertical distance Δz in each of the spatial distances, and ensure that each horizontal distance is equal, each vertical distance is equal, and each vertical distance is equal.

[0052] The distance between adjacent point x0-1 and the measuring point (x0, y0, z0) along the x-axis is the horizontal distance Δx, and the distance between adjacent point x0+1 and the measuring point (x0, y0, z0) along the x-axis is also the horizontal distance Δx. The distance between adjacent point y0-1 and the measuring point (x0, y0, z0) along the y-axis is the vertical distance Δy, and the distance between adjacent point y0+1 and the measuring point (x0, y0, z0) along the y-axis is also the vertical distance Δy. The distance between adjacent point z0-1 and the measuring point (x0, y0, z0) along the z-axis is the vertical distance Δz, and the distance between adjacent point z0+1 and the measuring point (x0, y0, z0) along the z-axis is also the vertical distance Δz.

[0053] S303, determine the horizontal neighbor, vertical neighbor and vertical neighbor among each of the adjacent points; determine the wind environment training data of the horizontal neighbor in the horizontal direction, the wind environment training data of the vertical neighbor in the vertical direction and the wind environment training data of the vertical neighbor in the vertical direction in the wind environment training data of the adjacent points.

[0054] The horizontal neighboring points include x0-1 and x0+1, the vertical neighboring points include y0-1 and y0+1, and the vertical adjacent points include z0-1 and z0+1.

[0055] The training data for wind environment of lateral neighbor points in the lateral direction includes u x0+1 and u x0-1 , where u x0+1 Let u be the training wind speed of the lateral neighbor x0+1 in the lateral direction. x0-1 Let x0-1 be the training wind speed in the lateral direction of the lateral neighbor point.

[0056] The longitudinal wind environment training data for the longitudinal neighbor points includes v y0+1 and v y0-1 , where v y0+1 Let v be the training wind speed of the longitudinal neighbor y0+1 in the longitudinal direction. y0-1 Let y0-1 be the training wind speed of the longitudinal neighbor point in the longitudinal direction.

[0057] The vertical neighboring points' wind environment training data in the vertical direction include w z0+1 and w z0-1 , where w z0+1 For the training wind speed of the vertical neighbor z0+1 in the vertical direction, w z0-1 Let z0-1 be the training wind speed of the vertical neighbor point z0-1 in the vertical direction.

[0058] In this embodiment, all training wind speeds are wind speeds output by the neural network model.

[0059] S304, based on the horizontal distance, the vertical distance, the longitudinal distance, the wind environment training data of the horizontal neighboring points in the horizontal direction, the wind environment training data of the longitudinal neighboring points in the longitudinal direction, and the wind environment training data of the vertical neighboring points in the vertical direction, determine the training rate of change Mass of the building wind environment training data. pre .

[0060] Masspre = (ux 0+1 -ux0-1)×(Δy×Δz)+(vy 0+1 -vy0-1)×(Δx×Δz)+(wz 0+1 -wz0-1)×(Δy×Δx)

[0061] In this embodiment, the true rate of change Mass in step S300 cal :

[0062]

[0063] In the formula, This represents the actual wind speed at the horizontal neighbor point x0+1 (that is, the actual wind speed at the horizontal direction corresponding to x0+1 in the building form sample matrix X). It represents the actual wind speed of the horizontal neighbor x0-1 in the horizontal direction (that is, the actual wind speed of x0-1 corresponding to the building form sample matrix X in the horizontal direction). This represents the true wind speed at the vertical neighbor point y0+1 in the vertical direction (that is, the true wind speed at y0+1 corresponding to the building form sample matrix X in the vertical direction). It represents the true wind speed of the vertical neighbor point y0-1 in the vertical direction (that is, the true wind speed of y0-1 corresponding to the building form sample matrix X in the vertical direction). This represents the true wind speed in the vertical direction at the vertical neighbor point z0+1 (that is, the true wind speed in the vertical direction corresponding to z0+1 in the building form sample matrix X). This represents the true wind speed of the vertical neighbor point z0-1 in the vertical direction (that is, the true wind speed of z0-1 corresponding to the building form sample matrix X in the vertical direction).

[0064] In this embodiment, the loss function is... data :

[0065]

[0066] n is the total number of sampling points within a cube formed by a given measuring point and all its adjacent points (where the sampling points include...). Figure 2 The measurement points and adjacent points within the cube, as well as other points within the cube, are included. Y represents the true wind speed at the k-th sampling point.k The training wind speed is the output of the kth sampling point of the neural network model.

[0067] The total loss function in this embodiment is:

[0068] loss=(1-u)lossdata+u(Masspre-Masscal)

[0069] In the formula, u is a hyperparameter less than 1.

[0070] Example 2, based on Example 1, provides a prediction method for predicting the wind environment of a building. This example includes the following specific steps: obtaining a building morphology matrix X′ and incident wind data, wherein the elements in the building morphology matrix are used to represent spatial location, and the values ​​of the elements are used to represent building attribute information or non-building attributes located at the spatial location, and the incident wind is the wind incident on the building environment; applying a building wind environment prediction model to the building morphology matrix to obtain building wind environment prediction data.

[0071] The incident wind data in this embodiment is 2 meters per second, which means that if the wind speed enters the environment where the building is located (such as a park with several buildings) at a speed of 2 meters per second, the wind speed at various locations in the environment is predicted.

[0072] The first step is to process the data within the calculation area using a 3m resolution raster. Centered on each raster point, samples are taken according to the data sampling frame (241, 241) to form a set of building form coding matrix data to be calculated (i.e., building form matrix X′).

[0073] The second step involves sequentially inputting the building form encoding matrix into the neural network model to obtain the corresponding building wind speed matrix (3,3,3,25). The first 3 represents the row number of X′, the second 3 represents the column number of X′, the third 3 represents the three wind directions, and 25 represents the 25 vertical sampling points. Points involving buildings will be skipped during execution.

[0074] The third step involves performing the calculations from step two. Each point to be calculated (i.e., the measuring point) will be calculated up to nine times. The multiple wind speed results calculated for each point are then summed and averaged according to the number of calculations to obtain the final wind speed distribution within the calculation range.

[0075] In summary, the model input of this invention is not limited to inputting only spatial location information, but can be replaced by inputting wind field information (building spatial morphology). After the model structure changes, the model can learn and predict the wind field for different building morphologies. The model output of this invention outputs the wind speed at multiple measurement points simultaneously. For the entire wind field, each measurement point will be calculated multiple times and the average value will be taken, which can improve the accuracy of model prediction. The model of this invention can also be used to calculate building acoustic environment and building thermal environment.

[0076] This embodiment also provides a building wind environment training device based on physical data hybrid driving, such as Figure 3 As shown, the device includes:

[0077] Dataset construction module 01 is used to construct a training dataset, which includes building form sample data and the corresponding real building wind environment data.

[0078] Training data generation module 02 is used to apply a neural network model to the building morphology sample data to obtain building wind environment training data output by the neural network model.

[0079] The loss function and rate of change calculation module 03 is used to determine the loss function of the neural network model based on the real building wind environment data and the building wind environment training data; determine the real rate of change of the real building wind environment data; and determine the training rate of change of the building wind environment training data.

[0080] Training module 04 is used to obtain the total loss function of the neural network model based on the actual rate of change, the training rate of change, and the loss function, and to iteratively train the neural network model based on the total loss function until the neural network model converges, thereby obtaining the building wind environment prediction model.

[0081] Based on the above embodiments, the present invention also provides a terminal device, the principle block diagram of which can be as follows: Figure 4 As shown, the terminal device includes a processor, memory, network interface, and display screen connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a physical data-driven building wind environment training method. The display screen can be an LCD screen or an e-ink screen.

[0082] Those skilled in the art will understand that Figure 4 The schematic diagram shown is only a partial structural diagram related to the present invention and does not constitute a limitation on the terminal device to which the present invention is applied. The specific terminal device may include more or fewer components than shown in the figure, or combine certain components, or have different component arrangements.

[0083] In one embodiment, a terminal device is provided, comprising a memory, a processor, and a physical data hybrid-driven building wind environment training method program stored in the memory and executable on the processor. When the processor executes the physical data hybrid-driven building wind environment training method program, it implements the following operation instructions:

[0084] Construct a training dataset, which includes building form sample data and the corresponding real building wind environment data;

[0085] A neural network model is applied to the architectural form sample data to obtain architectural wind environment training data output by the neural network model.

[0086] Based on the real building wind environment data and the building wind environment training data, determine the loss function of the neural network model; determine the true rate of change of the real building wind environment data; determine the training rate of change of the building wind environment training data;

[0087] Based on the actual rate of change, the training rate of change, and the loss function, the total loss function of the neural network model is obtained. Based on the total loss function, the neural network model is iteratively trained until the neural network model converges, thus obtaining the building wind environment prediction model.

[0088] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided by this invention can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for rapid wind environment prediction based on a physically consistent neural network, characterized in that, include: Construct a training dataset, which includes building form sample data and the corresponding real building wind environment data; A neural network model is applied to the architectural form sample data to obtain architectural wind environment training data output by the neural network model. Based on the real building wind environment data and the building wind environment training data, determine the loss function of the neural network model; determine the true rate of change of the real building wind environment data; Determine the training rate of change of the building wind environment training data; Based on the actual rate of change, the training rate of change, and the loss function, the total loss function of the neural network model is obtained. Based on the total loss function, the neural network model is iteratively trained until the neural network model converges, thus obtaining the building wind environment prediction model. Obtain building form matrix and incident wind data. The elements in the building form matrix are used to represent spatial location, and the values ​​of the elements are used to represent building attribute information or non-building attributes located at the spatial location. Incident wind is the wind that is incident on the building environment. The building wind environment prediction model is applied to the building form matrix to obtain building wind environment prediction data, that is, when the incident wind data enters the building environment, the wind speed at each location point in the environment is predicted. Calculating the training rate of change involves the following specific steps: Determine the adjacent point wind environment training data of each adjacent point in the building wind environment training data; Obtain the spatial distance between the measuring point and each of the adjacent points, where each of the adjacent points is a point adjacent to the measuring point, and the measuring point is a location that is not a building. Each horizontal distance, each vertical distance, and each elongated distance in the aforementioned spatial distances are determined, and each horizontal distance is equal to the others, each vertical distance is equal to the others, and each elongated distance is equal to the others. Determine the lateral, longitudinal, and vertical neighboring points among each of the adjacent points; determine the wind environment training data of the lateral neighboring points in the lateral direction, the wind environment training data of the longitudinal neighboring points in the longitudinal direction, and the wind environment training data of the vertical neighboring points in the vertical direction in the wind environment training data of the adjacent points. Based on the horizontal distance, the vertical distance, the longitudinal distance, the wind environment training data of the horizontal neighbor points in the horizontal direction, the wind environment training data of the longitudinal neighbor points in the longitudinal direction, and the wind environment training data of the vertical neighbor points in the vertical direction, the training rate of change of the building wind environment training data is determined.

2. The method for rapid wind environment prediction based on a physically consistent neural network as described in claim 1, characterized in that, Training variation rate : In the formula, The spatial location of the measuring point. , , , , , These are the positions of six adjacent points. Horizontal neighbor Training wind speed in the lateral direction, Horizontal neighbor Training wind speed in the lateral direction, Vertical neighbor Training wind speed in the longitudinal direction, Vertical neighbor Training wind speed in the longitudinal direction, Vertical neighbor Training wind speed in the vertical direction, Vertical neighbor Training wind speed in the vertical direction, adjacent points With measuring points Lateral distance along the x-axis, adjacent points With measuring points The vertical distance along the y-axis, adjacent points With measuring points Vertical distance along the z-axis.

3. The method for rapid wind environment prediction based on a physically consistent neural network as described in claim 1 or 2, characterized in that, The true rate of change : In the formula, Horizontal neighbor The actual wind speed in the lateral direction Horizontal neighbor The actual wind speed in the lateral direction Vertical neighbor The actual wind speed in the longitudinal direction, Vertical neighbor The actual wind speed in the longitudinal direction, Vertical neighbor The actual wind speed in the vertical direction, Vertical neighbor The actual wind speed in the vertical direction.

4. The method for rapid wind environment prediction based on a physically consistent neural network as described in claim 1, characterized in that, Also includes: Incident wind data is incorporated into the building's wind environment to predict wind speeds at various locations within that environment.

5. The method for rapid wind environment prediction based on a physically consistent neural network as described in claim 2, characterized in that, Also includes: Computational Fluid Dynamics (CFD) is applied to urban building morphology cases to obtain real-world wind environment data for the buildings. These urban building morphology cases include three-dimensional data of building morphology at various locations in the city and wind speeds at different heights at each location.

Images