Wind field prediction method and device for group building, equipment and medium

Abstract

The invention relates to the technical field of wind field prediction, and provides a wind field prediction method and device for group buildings, equipment and a medium. The method comprises the following steps: acquiring positions of a plurality of buildings in a plurality of target areas and a wind field cloud picture of each target area; constructing a comprehensive loss function of the generative adversarial network according to the wind field cloud picture of each target area; training the generative adversarial network by using the comprehensive loss function to obtain a trained generative adversarial network; and performing wind field prediction on the to-be-predicted region by using the trained generative adversarial network to obtain a final wind field prediction cloud picture of the to-be-predicted region. According to the method, the wind field prediction accuracy of the group building can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of wind field prediction, and in particular to a method, device, equipment and medium for predicting a wind field for a group of buildings. Background Art

[0002] The wind loads on a building when disturbed by surrounding structures differ significantly from those on an isolated structure. This phenomenon is known as the interference effect. Research has shown that the impact of the interference effect can be several or even dozens of times greater than the size of the interfering building itself. Due to the interference effect, urban wind fields in modern cities, influenced by dense clusters of high-rise buildings, are often extremely complex and difficult to predict. This can lead to numerous issues, including drone yaw, occupant comfort, and even loss of life and property. Accurately predicting the wind field around buildings in the presence of the interference effect is crucial to addressing these challenges.

[0003] Generally, the more buildings in a complex, the more complex and difficult to predict the wind field due to interference effects. Over the past few decades, numerous researchers have dedicated themselves to uncovering the mechanisms of interference effects in order to understand the patterns of wind fields in complexes and, ultimately, predict them. However, due to time and cost constraints, existing research has mostly focused on the interference effects of two or three buildings, and the resulting conclusions are only applicable to predicting the flow field for complexes containing only two or three buildings. In contrast, systematic research on wind fields in complexes of four or more buildings is very limited. This is because the number of building layouts that must be considered increases exponentially with each additional building, leading to a surge in the number of test conditions required, and a sharp increase in the time and financial costs. Furthermore, as the number of buildings increases, the patterns of wind fields in complexes become increasingly complex, and simple parameters (such as building spacing and wind direction angle) are no longer sufficient to accurately describe the wind field patterns in complexes containing multiple buildings.

[0004] Therefore, it can be said that due to time, funding, and the inherent complexity of the wind field of a group of buildings, the current research model based on a large number of wind tunnel tests to summarize the interference effects of group buildings is almost impossible to generalize to multiple buildings. This is also why, over the past few decades, existing research has been limited to the "initial stage" of analyzing the interference effects of only two or three buildings, and research on four or more buildings is very rare. The few studies that have focused on groups of four or more buildings have mainly focused on certain specific layouts (such as serial or square arrangements). Their research results have very limited guidance for the wind-resistant design of group buildings considering interference effects.

[0005] In summary, at the current stage, the academic community has a serious lack of understanding of the interference effect of group buildings, which leads to the problem of low accuracy of wind field prediction for group buildings in existing wind field prediction methods. Summary of the Invention

[0006] The present application provides a method, device, equipment and medium for predicting the wind field of a group of buildings, which can solve the problem of low accuracy in wind field prediction for a group of buildings.

[0007] In a first aspect, an embodiment of the present application provides a method for predicting a wind field for a group of buildings, the method comprising:

[0008] Obtain the locations of multiple buildings in multiple target areas and the wind field cloud map of each target area; the wind field cloud map is used to describe the wind pressure value data of the target area;

[0009] A comprehensive loss function for the generative adversarial network is constructed based on the wind field cloud map of each target area. The comprehensive loss function is used to describe the accuracy of the predicted wind field cloud map of each target area generated by the generative adversarial network under the conditions of building location, brightness, and wind field wake vortex.

[0010] The generative adversarial network is trained using a comprehensive loss function to obtain a trained generative adversarial network.

[0011] The trained generative adversarial network is used to predict the wind field in the prediction area and obtain the final wind field prediction cloud map of the prediction area.

[0012] Optionally, a comprehensive loss function for the generative adversarial network is constructed based on the wind field cloud map of each target area, including:

[0013] Construct true and false loss function, hidden layer loss function, wake vortex gradient loss function, and empirical mode feature loss function based on all wind field cloud images;

[0014] Construct a building location loss function based on the locations of all buildings;

[0015] Calculate the average brightness of each wind field cloud image and construct a brightness loss function based on all average brightness;

[0016] The true and false loss function, hidden layer loss function, wake vortex gradient loss function, empirical mode feature loss function, building position loss function and brightness loss function are combined to obtain a comprehensive loss function.

[0017] Optionally, the true / false loss function is:

[0018]

[0019]

[0020] in, Represents the true and false loss function value of the generator, Represents the true and false loss function value of the discriminator, p z(z) represents the set of all predicted wind field cloud images, G(z) represents the z-th predicted wind field cloud image, D1(G(z)) represents the output of the discriminator for the z-th predicted wind field cloud image, ||||2 represents the L2 norm, p data (x) represents the set of all wind field cloud images, and D1(x) represents the output of the discriminator for the wind field cloud image x;

[0021] The hidden layer loss function is:

[0022]

[0023] Among them, L layer Represents the hidden layer loss function value, o i (D1(x)) represents the hidden layer output obtained by processing the wind field cloud image x through the i-th hidden layer, o i (D1(G(z))) represents the hidden layer output obtained by processing the z-th predicted wind field cloud image by the i-th hidden layer, and k represents the number of hidden layers;

[0024] The wake vortex gradient loss function is:

[0025]

[0026] Among them, L grad represents the value of the wake vortex gradient loss function, represents the vortex structure gradient of the wind field cloud map x, Represents the vortex structure gradient of the z-th predicted wind field cloud map.

[0027] Optionally, the empirical mode feature loss function is:

[0028]

[0029] Among them, L mode represents the empirical mode feature loss function value, represents the nth-order empirical mode information of the wind field cloud image x, represents the nth-order empirical modal information of the zth predicted wind field cloud map, and m represents the order of the empirical modal information.

[0030] Optionally, the building location loss function is:

[0031]

[0032] Among them, L position represents the building location loss function value, D2(G(z)) represents all building locations in the z-th predicted wind field cloud map, and z * Represents the locations of all buildings in the target area corresponding to the z-th predicted wind field cloud map.

[0033] Optionally, the brightness loss function is:

[0034]

[0035] Among them, L bright represents the brightness loss function value, c(D1(x)) represents the average brightness of the wind field cloud image x, and c(D1(G(z))) represents the average brightness of the z-th predicted wind field cloud image.

[0036] Optionally, the comprehensive loss function is:

[0037]

[0038] Among them, L G Represents the comprehensive loss function value of the generator, L D Represents the comprehensive loss function value of the discriminator.

[0039] In a second aspect, the present application provides a wind field prediction device for a group of buildings, comprising:

[0040] An acquisition module is used to obtain the locations of multiple buildings in multiple target areas and a wind field cloud map of each target area; the wind field cloud map is used to describe the wind pressure value data of the target area;

[0041] The construction module constructs a comprehensive loss function for the generative adversarial network based on the wind field cloud map of each target area. The comprehensive loss function is used to describe the accuracy of the predicted wind field cloud map of each target area generated by the generative adversarial network under the conditions of building location, brightness, and wind field wake vortex.

[0042] The training module uses the comprehensive loss function to train the generative adversarial network to obtain the trained generative adversarial network;

[0043] The wind field prediction module uses the trained generative adversarial network to predict the wind field in the prediction area and obtains the final wind field prediction cloud map of the prediction area.

[0044] 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 above-mentioned wind field prediction method for a group of buildings when executing the above-mentioned computer program.

[0045] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing a computer program, which implements the above-mentioned wind field prediction method for a group of buildings when executed by a processor.

[0046] The above solution of the present application has the following beneficial effects:

[0047] In an embodiment of the present application, by obtaining the positions of multiple buildings in multiple target areas and the wind field cloud map of each target area, and then constructing a comprehensive loss function of a generative adversarial network based on the wind field cloud map of each target area, the generative adversarial network is trained using the comprehensive loss function to obtain a trained generative adversarial network, and finally the trained generative adversarial network is used to predict the wind field in the predicted area to obtain the final wind field prediction cloud map of the predicted area. Among them, the use of a generative adversarial network to generate a predicted wind field cloud map can describe the wind field information of an area with multiple buildings and realize the wind field prediction of a group of buildings. The constructed comprehensive loss function fully considers the characteristics of the wind field cloud map and can describe the accuracy of the predicted wind field cloud map in terms of brightness, wake vortex, and building position, effectively improving the effectiveness and accuracy of the comprehensive loss function. The accurate comprehensive loss function is used to train the generative adversarial network so that the performance of the generative adversarial network meets expectations. The high-performance generative adversarial network is used to obtain the final wind field prediction cloud map, which improves the accuracy of wind field prediction while realizing the wind field prediction of a group of buildings.

[0048] In addition, the wind field information of the group of buildings is described using wind field cloud maps, which realizes the conversion of the wind field prediction problem of the group of buildings into the prediction problem of wind field cloud maps. The wind field prediction of the group of buildings is realized by studying the two-dimensional images, and there is no need to perform complex wind field-related parameter analysis on the wind field information of the group of buildings, thus avoiding errors in the parameter analysis process.

[0049] Other beneficial effects of the present application will be described in detail in the subsequent specific implementation section. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] 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. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0051] Figure 1 A flow chart of a method for predicting a wind field for a group of buildings provided in one embodiment of the present application;

[0052] Figure 2 A schematic diagram of a two-dimensional steady-state simulation provided in one embodiment of the present application;

[0053] Figure 3 A schematic diagram of a wind field cloud map provided in one embodiment of the present application;

[0054] Figure 4 A schematic diagram of the structure of a generative adversarial network provided in one embodiment of the present application;

[0055] Figure 5 A mean error curve diagram provided in an embodiment of the present application;

[0056] Figure 6 A schematic diagram of predicted wind field cloud maps for different training cycles provided in an embodiment of the present application;

[0057] Figure 7 A schematic diagram of prediction accuracy provided in one embodiment of the present application;

[0058] Figure 8 A schematic diagram of the structure of a wind field prediction device for a group of buildings provided in one embodiment of the present application;

[0059] Figure 9 A schematic diagram of the structure of a terminal device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0060] 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.

[0061] 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.

[0062] 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.

[0063] As used in this specification and the appended claims, the term "if" can be interpreted as "when" or "upon" or "in response to determining" or "in response to detecting," depending on the context. Similarly, the phrase "if it is determined" or "if [described condition or event] is detected" can be interpreted as meaning "upon determination" or "in response to determining" or "upon detection of [described condition or event]" or "in response to detecting [described condition or event]," depending on the context.

[0064] 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.

[0065] References to "one embodiment" or "some embodiments" in this specification mean that a particular feature, structure, or characteristic described in conjunction with that 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 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. The terms "including," "comprising," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0066] In response to the problem of low accuracy of existing wind field predictions for group buildings, an embodiment of the present application provides a wind field prediction method for group buildings. The wind field prediction method obtains the positions of multiple buildings in multiple target areas and the wind field cloud map of each target area, and then constructs a comprehensive loss function of a generative adversarial network based on the wind field cloud map of each target area. The comprehensive loss function is then used to train the generative adversarial network to obtain a trained generative adversarial network. Finally, the trained generative adversarial network is used to perform wind field prediction on the area to be predicted to obtain the final wind field prediction cloud map of the area to be predicted. Among them, the generative adversarial network is used to generate the predicted wind field cloud map, which can describe the wind field information of the area with multiple buildings and realize the wind field prediction of the group of buildings. The constructed comprehensive loss function fully takes into account the characteristics of the wind field cloud map, and can describe the accuracy of the predicted wind field cloud map in terms of brightness, wake vortex, and building location, effectively improving the effectiveness and accuracy of the comprehensive loss function. The accurate comprehensive loss function is used to train the generative adversarial network, so that the performance of the generative adversarial network meets the expectations. The high-performance generative adversarial network is used to obtain the final wind field prediction cloud map, which improves the accuracy of wind field prediction while realizing the wind field prediction of the group of buildings.

[0067] In addition, the wind field information of the group of buildings is described using wind field cloud maps, which realizes the conversion of the wind field prediction problem of the group of buildings into the prediction problem of wind field cloud maps. The wind field prediction of the group of buildings is realized by studying the two-dimensional images, and there is no need to perform complex wind field-related parameter analysis on the wind field information of the group of buildings, thus avoiding errors in the parameter analysis process.

[0068] Next, the wind field prediction method for a group of buildings provided in this application is exemplified.

[0069] like Figure 1 As shown, the wind field prediction method for a group of buildings provided in this application includes the following steps:

[0070] Step 11: Obtain the locations of multiple buildings in multiple target areas and a wind field cloud map of each target area.

[0071] The wind field cloud map is used to describe the wind pressure value data of the target area. The pixels in the wind field cloud map include pixels used to describe buildings and pixels used to describe wind pressure values ​​(the pixel value of the pixel corresponds to the wind pressure value). The target area is an area with multiple buildings, such as a residential area.

[0072] In some embodiments of the present application, the wind field cloud map and the building position can be obtained by simulating the target area (the relative position of the building in the simulation model of the target area, such as the target area in the simulation software is a 60D×30D rectangle, and a two-dimensional coordinate system is constructed with the lower left corner vertex of the rectangle as the circle point, and the building position of a building is a square area with (30D, 15D) as the center and a side length of 1D). For example, a two-dimensional steady-state simulation can be performed using a turbulence model to obtain a wind field cloud map for each target area, and the position of the building can be obtained using a self-programming script.

[0073] Exemplarily, the above two-dimensional steady-state simulation is performed by ANSYS Fluent software, such as Figure 2 As shown, the horizontal direction is the x-axis and the vertical direction is the y-axis, where the inlet boundary is set as the velocity inlet, the average wind speed v = 8m / s, and the turbulence intensity is 5%; the outlet boundary is set as the pressure outlet, and the two sides are set as symmetry boundaries. During the simulation, the building size in each area is uniformly set to D = 0.1m in length and width, and the calculation domain size is 60D×30D. The number of buildings in each area is generated by a self-programmed random number generator, and the number of buildings is set between 2-5. The position of the buildings in the area is controlled by another self-programmed script: one of the buildings is located at the center of the calculation domain, with the center coordinates of (30D, 15D), and the remaining buildings are randomly distributed in a smaller area around the center of the calculation domain, with a size of 20D×10D. Figure 2 Building 1 is located at the center of the computational domain, and buildings 2 and 3 are located in a 20D×10D area centered on building 1.

[0074] The simulated wind field cloud map is as follows Figure 3 As shown, the wind field cloud map is cropped and resampled according to the size of 20D×30D to obtain the final wind field cloud map, where the pixel value of the pixel point represents the wind pressure value C p , white squares represent buildings.

[0075] It is worth mentioning that by obtaining the wind field cloud map of the target area, the steady-state wind field prediction for the group of buildings is equivalently converted into the steady-state wind field cloud map prediction for the group of buildings, providing effective data for wind field prediction for the group of buildings.

[0076] Step 12: Construct a comprehensive loss function of the generative adversarial network based on the wind field cloud map of each target area.

[0077] The predicted wind cloud map above is used to describe the locations of all buildings within the generated target area and the predicted wind pressure data. A comprehensive loss function is used to describe the accuracy of the predicted wind cloud map generated by the GAN for each target area, considering building locations, brightness, and wind wake vortices.

[0078] In some embodiments of the present application, the above-mentioned generative adversarial network is a generative adversarial network combined with an attention mechanism, such as Figure 4 As shown, the generative adversarial network includes the generator Figure 4 a and the discriminator Figure 4 b. The generator consists of a 5-layer Unet model, P i is the input of the generator (i.e., the positions of all buildings in the target area), the solid arrows represent the fully connected layers, the horizontal filled blocks and the white blocks represent the feature maps, the downward black arrows represent the two-dimensional average pooling 2×2, the upward black arrows represent the two-dimensional transposed convolution 2×2, the white arrows represent the double convolution 3×3, the oblique filled arrows represent the self-attention, the horizontal filled arrows represent the convolution 1×1, the dotted arrows represent the copy, and the grid filled arrows represent the residual blocks. The discriminator receives the predicted wind field cloud map and the wind field cloud map output by the generator and processes the predicted wind field cloud map and the wind field cloud map. The discriminator has two branches. One branch processes the real image (i.e., the wind field cloud map) and the forged image (i.e., the predicted wind field cloud map) through three 4×4 convolution blocks connected in sequence to obtain Q3 (the building position P in the predicted wind field cloud map). i ), another branch, including two 1×1 convolution blocks and four 4×4 convolution blocks connected in sequence, performs multiple convolution sampling (the black triangle in the figure represents a 1×1 convolution block, and the white triangle represents a 4×4 convolution block), processes the input data, and outputs the second 1×1 convolution block and the layer output Q1 of each 4×4 convolution block. Then, a fully connected layer is used to predict the real image and the forged image as real or fake Q2.

[0079] It should be noted that after the generator generates an image based on the input data, the discriminator will determine the difference between the image and the training image (i.e., the wind field cloud map), and prompt the generator to continuously improve the generated image so that the image generated by the generator is close to the wind field cloud map, and finally obtain the predicted wind field cloud map.

[0080] The above comprehensive loss function is used to describe the accuracy of all predicted wind field cloud maps under the conditions of building location, brightness, and wind field wake vortex. For building location, the predicted wind field cloud map includes a description of the location of all buildings in the target area. By analyzing the difference between the building location in the predicted wind field cloud map and the building location in the wind field cloud map, the accuracy of the predicted wind field cloud map can be analyzed. For brightness, the pixel value in the predicted wind field cloud map reflects the wind pressure value in the corresponding area of ​​the pixel. By analyzing the difference between the pixel brightness of the predicted wind field cloud map and the pixel brightness of the wind field cloud map, the accuracy of the predicted wind field cloud map can be analyzed. For wake vortex, turbulent wind will form a vortex downstream when passing through a building. The topological structure of this vortex has several inherent characteristics: the vortex shape is circular or elliptical, the closer to the center of the vortex, the greater the negative pressure, and the vortex has its own specific rotation direction. By analyzing the difference between the wake vortex in the predicted wind field cloud map and the wake vortex in the wind field cloud map, the accuracy of the predicted wind field cloud map can be analyzed.

[0081] In some embodiments of the present application, the step of constructing a comprehensive loss function of a generative adversarial network based on the wind field cloud map of each target area is specifically as follows:

[0082] In the first step, the true and false loss function, hidden layer loss function, wake vortex gradient loss function, and empirical mode feature loss function are constructed based on all wind field cloud images.

[0083] Specifically, the true and false loss function is:

[0084]

[0085] in, Represents the true and false loss function value of the generator, Represents the true and false loss function value of the discriminator, p z (z) represents the set of all predicted wind field cloud images, G(z) represents the z-th predicted wind field cloud image, D1(G(z)) represents the output of the discriminator for the z-th predicted wind field cloud image, ||||2 represents the L2 norm, p data (x) represents the set of all wind field cloud images, and D1(x) represents the output of the discriminator for the wind field cloud image x.

[0086] The hidden layer loss function is:

[0087]

[0088] Among them, L layer Represents the hidden layer loss function value, o i (D1(x)) represents the hidden layer output obtained by processing the wind field cloud image x through the i-th hidden layer, o i(D1(G(z))) represents the hidden layer output obtained by processing the z-th predicted wind field cloud image through the i-th hidden layer, and k represents the number of hidden layers.

[0089] It should be noted that the above hidden layer is the structure of the discriminator of the generative adversarial network, such as Figure 4 In the branch where the discriminator determines whether the predicted wind field cloud image is real or false, multiple hidden layers are the second 1×1 convolution block and all 4×4 convolution layers.

[0090] The wake vortex gradient loss function is:

[0091]

[0092] Among them, L grad represents the value of the wake vortex gradient loss function, represents the vortex structure gradient of the wind field cloud map x, Represents the vortex structure gradient of the z-th predicted wind field cloud map.

[0093] The empirical mode feature loss function is:

[0094]

[0095] Among them, L mode represents the empirical mode feature loss function value, represents the nth-order empirical mode information of the wind field cloud image x, represents the nth-order empirical mode information of the zth predicted wind field cloud map, m represents the order of the empirical mode information, and the two-dimensional empirical mode decomposition method can be used to obtain the multi-order empirical mode information of the wind field cloud map and the predicted wind field cloud map.

[0096] It's important to note that turbulent winds passing over buildings can form vortices downstream. The topological structure of these vortices exhibits several inherent characteristics, such as a circular or elliptical shape, increasing negative pressure closer to the vortex center, and a specific rotational direction. Generally speaking, the pressure gradient of a vortex structure should be smooth and continuous. The wake vortex gradient loss function aims to deepen the generative adversarial network's understanding of the vortex topology results, addressing the issue of localized pixelation in generated images. It also improves the continuity and smoothness of the vortex structure gradients in the generated images, enhancing the realism of the images.

[0097] In the second step, a building location loss function is constructed based on the locations of all buildings.

[0098] Specifically, the building position loss function is:

[0099]

[0100] Among them, L positionrepresents the building location loss function value, D2(G(z)) represents all building locations in the z-th predicted wind field cloud map, and z * Represents the locations of all buildings in the target area corresponding to the z-th predicted wind field cloud map.

[0101] The third step is to calculate the average brightness of each wind field cloud image and construct a brightness loss function based on all average brightness.

[0102] Specifically, the brightness loss function is:

[0103]

[0104] Among them, L bright represents the brightness loss function value, c(D1(x)) represents the average brightness of the wind field cloud image x, and c(D1(G(z))) represents the average brightness of the z-th predicted wind field cloud image.

[0105] In the fourth step, the true and false loss function, hidden layer loss function, wake vortex gradient loss function, empirical mode feature loss function, building position loss function and brightness loss function are combined to obtain a comprehensive loss function.

[0106] Specifically, the comprehensive loss function is:

[0107]

[0108] Among them, L G Represents the comprehensive loss function value of the generator, L D Represents the comprehensive loss function value of the discriminator.

[0109] It should be noted that the above generators and discriminators are both generators and discriminators in generative adversarial networks.

[0110] It is worth mentioning that the constructed comprehensive loss function fully takes into account the characteristics of the wind field cloud map, and can describe the accuracy of the predicted wind field cloud map in three aspects: brightness, wake vortex, and building location, effectively improving the effectiveness and accuracy of the comprehensive loss function.

[0111] Step 13: Use the comprehensive loss function to train the generative adversarial network to obtain a trained generative adversarial network.

[0112] Specifically, according to the wind field cloud maps of all target areas obtained in step 11 and the predicted wind field cloud maps of each target area generated by the generative adversarial network, the value of the comprehensive loss function is calculated, and it is determined whether the value of the comprehensive loss function is less than the preset loss value to train the generative adversarial network. If the comprehensive loss function value of the generator and the comprehensive loss function value of the discriminator are both less than the preset loss value, the generative adversarial network is considered to be a trained generative adversarial network. If only one of the comprehensive loss function value of the generator and the comprehensive loss function value of the discriminator is greater than or equal to the preset loss value, the parameters of the generative adversarial network are adjusted.

[0113] For example, the generative adversarial network can also be trained using the Adam optimizer, with a total of 200 rounds of iterative training using the Adam optimizer. To stabilize the training process, the strategy for setting the learning rate during training is as follows: the initial learning rate of the generator is set to 1e-4, and the initial learning rate of the discriminator is set to 4e-4. After each round, the learning rate is multiplied by 0.95. In addition, the learning rate is reset to the initial value every 50 rounds. During the training process, the generator and discriminator are trained alternately. While training the generator, the parameters of the discriminator remain unchanged, and vice versa. In each iteration, we update the parameters of the discriminator once and the parameters of the generator twice. This approach ensures that the discriminator and generator progress at the same rate.

[0114] It is worth mentioning that the accurate comprehensive loss function is used to train the generative adversarial network, so that the performance of the generative adversarial network meets expectations.

[0115] The above training is illustrated below with reference to a specific example.

[0116] The mean error of the predicted wind field cloud map is as follows: Figure 5 As shown, Figure 5 The horizontal axis represents the training cycle, the vertical axis represents the size of the mean error, and the learning rate is reset every 50 training cycles. Figure 5 As can be seen from the figure, the mean error decreases rapidly in the initial period and then gradually stabilizes with a minimum value of about 0.04. Since the learning rate is reset every 50 rounds, Figure 5 Two local peaks can be observed in the residual curve of . Figure 5 There are two local peaks on the residual curve at epochs 50 and 100. These local peaks disappear quickly with the increase in the number of training times, indicating that the training strategy is reliable.

[0117] The predicted wind field cloud maps of different training cycles are as follows Figure 6 As shown, Figure 6 a is the predicted wind field cloud map when the training cycle is 1, Figure 6 b is the predicted wind field cloud map when the training cycle is 5, Figure 6c is the predicted wind field cloud map with a training period of 20 hours, Figure 6 d is the predicted wind field cloud map when the training cycle is 50, Figure 6 e is the predicted wind field cloud map when the training period is 200, Figure 6 f is the CFD result (i.e., the predicted wind field cloud map at the end of training).

[0118] Step 14: Use the trained generative adversarial network to perform wind field prediction on the prediction area to obtain the final wind field prediction cloud map of the prediction area.

[0119] The above-mentioned area to be predicted is the area where wind field prediction is required, and the area includes at least two buildings. The final wind field prediction cloud map is used to describe the locations and wind pressure value data of all buildings in the area to be predicted.

[0120] Specifically, the location information of all buildings in the area to be predicted is obtained, and all the location information is input into the trained generative adversarial network to generate the final wind field prediction cloud map of the area to be predicted.

[0121] It is worth mentioning that the use of generative adversarial networks to generate predicted wind field cloud maps can describe the wind field information of areas with multiple buildings and realize wind field prediction for groups of buildings. The constructed comprehensive loss function fully takes into account the characteristics of the wind field cloud map and can describe the accuracy of the predicted wind field cloud map in terms of brightness, wake vortex, and building location, effectively improving the effectiveness and accuracy of the comprehensive loss function. The accurate comprehensive loss function is used to train the generative adversarial network, so that the performance of the generative adversarial network meets expectations. The high-performance generative adversarial network is used to obtain the final wind field prediction cloud map, which improves the accuracy of wind field prediction while realizing wind field prediction for groups of buildings.

[0122] The wind field prediction method for a group of buildings of the present application is exemplified below with reference to a specific example.

[0123] like Figure 7 As shown in the figure, the horizontal axis represents the actual wind pressure value of the wind field cloud map, and the vertical axis represents the predicted wind pressure value of the predicted wind field cloud map. Figure 7 a is a schematic diagram showing the wind field prediction accuracy of an area with two buildings using the method of this application, and the determination coefficient R 2 =0.98, Figure 7 b is a schematic diagram showing the wind prediction accuracy for an area with three buildings, with the coefficient of determination R 2 =0.98, Figure 7 c represents the prediction accuracy diagram of the wind field prediction for an area with four buildings, and the determination coefficient R 2 =0.96, Figure 7d represents the wind prediction accuracy diagram for an area with five buildings, and the coefficient of determination R 2 =0.91.

[0124] It can be seen that the prediction accuracy of the method of this application is extremely high when predicting the wind field with only two and three buildings. 2 As high as 0.98, the prediction accuracy decreases with the increase of the number of buildings, but R 2 The lowest is still above 0.90.

[0125] In addition, compared with traditional methods (wind tunnel experiments / numerical simulations) that usually take several hours or even days to obtain building wind field information, this application only takes a few seconds to obtain the wind field information of the area to be predicted after obtaining the trained generative adversarial network, effectively improving the efficiency of wind field prediction for group buildings.

[0126] This application successfully applies generative adversarial networks to the study of wind patterns in clusters of buildings, proposing a new research paradigm that significantly improves the limitations of traditional methods, which can only study wind patterns between two or three buildings. It successfully predicts wind patterns for clusters of up to five buildings in any arrangement with high accuracy. Theoretically, applying this research paradigm, by constructing a training set containing a larger number of buildings in step 11 and continuously increasing the complexity of the generative adversarial network, it is possible to predict wind patterns for clusters of any number and arrangement of buildings.

[0127] The wind field prediction device for a group of buildings provided in this application is exemplarily described below.

[0128] like Figure 8 As shown, an embodiment of the present application provides a wind field prediction device for a group of buildings. The wind field prediction device 800 includes:

[0129] Acquisition module 801, acquires the locations of multiple buildings in multiple target areas and a wind field cloud map of each target area; the wind field cloud map is used to describe the wind pressure value data of the target area;

[0130] Construction module 802 constructs a comprehensive loss function for the generative adversarial network based on the wind field cloud map of each target area. The comprehensive loss function is used to describe the accuracy of the predicted wind field cloud map of each target area generated by the generative adversarial network under the conditions of building location, brightness, and wind field wake vortex.

[0131] A training module 803 trains the generative adversarial network using a comprehensive loss function to obtain a trained generative adversarial network;

[0132] The wind field prediction module 804 uses the trained generative adversarial network to perform wind field prediction on the prediction area to obtain a final wind field prediction cloud map of the prediction area.

[0133] 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.

[0134] 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.

[0135] like Figure 9 As shown, an embodiment of the present application provides a terminal device, and the terminal device D10 of this embodiment includes: at least one processor D100 ( Figure 9 Only one processor is shown in the figure), a memory D101, and a computer program D102 stored in the memory D101 and executable on the at least one processor D100, wherein the processor D100 implements the steps of any of the above method embodiments when executing the computer program D102.

[0136] Specifically, when the processor D100 executes the computer program D102, it obtains the positions of multiple buildings in multiple target areas and the wind field cloud map of each target area, and then constructs a comprehensive loss function of the generative adversarial network based on the wind field cloud map of each target area, and then uses the comprehensive loss function to train the generative adversarial network to obtain the trained generative adversarial network, and finally uses the trained generative adversarial network to predict the wind field in the prediction area to obtain the final wind field prediction cloud map of the prediction area. Among them, the use of the generative adversarial network to generate the predicted wind field cloud map can describe the wind field information of the area with multiple buildings and realize the wind field prediction of the group of buildings. The constructed comprehensive loss function fully considers the characteristics of the wind field cloud map and can describe the accuracy of the predicted wind field cloud map in three aspects: brightness, wake vortex, and building location, effectively improving the effectiveness and accuracy of the comprehensive loss function. The accurate comprehensive loss function is used to train the generative adversarial network so that the performance of the generative adversarial network reaches the expected level. The high-performance generative adversarial network is used to obtain the final wind field prediction cloud map, which improves the accuracy of the wind field prediction while realizing the wind field prediction of the group of buildings.

[0137] In addition, the wind field information of the group of buildings is described using wind field cloud maps, which realizes the conversion of the wind field prediction problem of the group of buildings into the prediction problem of wind field cloud maps. The wind field prediction of the group of buildings is realized by studying the two-dimensional images, and there is no need to perform complex wind field-related parameter analysis on the wind field information of the group of buildings, thus avoiding errors in the parameter analysis process.

[0138] The processor D100 may be a central processing unit (CPU), or may be another general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (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.

[0139] In some embodiments, the memory D101 may be an internal storage unit of the terminal device D10, such as a hard disk or memory of the terminal device D10. In other embodiments, the memory D101 may also be an external storage device of the terminal device D10, such as a plug-in hard disk, a smart memory card (SMC, SmartMedia Card), a secure digital (SD, Secure Digital) card, a flash card, etc. equipped on the terminal device D10. Furthermore, the memory D101 may also include both an internal storage unit of the terminal device D10 and an external storage device. The memory D101 is used to store an operating system, an application program, a boot loader (BootLoader), data, and other programs, such as the program code of the computer program. The memory D101 may also be used to temporarily store data that has been output or is to be output.

[0140] 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.

[0141] 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.

[0142] If the integrated unit is implemented in the form of 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 processes in the above-mentioned embodiment method 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 that can carry the computer program code to the wind field prediction method device / terminal device for group buildings, 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 disk, magnetic disk or optical disk. In some jurisdictions, according to legislation and patent practice, computer-readable media cannot be electric carrier signals and telecommunication signals.

[0143] 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.

[0144] 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.

[0145] The above is a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles described in the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A method for predicting wind fields for a group of buildings, characterized in that: include: Acquire the locations of multiple buildings in multiple target areas and a wind field cloud map of each target area; The wind field cloud map is used to describe the wind pressure value data of the target area; A comprehensive loss function of a generative adversarial network is constructed according to the wind field cloud map of each target area; the comprehensive loss function is used to describe the accuracy of the predicted wind field cloud map of each target area generated by the generative adversarial network under the building location, brightness, and wind field wake vortex; Using the comprehensive loss function to train the generative adversarial network to obtain a trained generative adversarial network; The trained generative adversarial network is used to perform wind field prediction on the area to be predicted, and a final wind field prediction cloud map of the area to be predicted is obtained.

2. The wind field prediction method according to claim 1, characterized in that: The comprehensive loss function of the generative adversarial network is constructed according to the wind field cloud map of each target area, including: According to all wind field cloud images, the true and false loss function, hidden layer loss function, wake vortex gradient loss function and empirical mode feature loss function are constructed; Construct a building location loss function based on the locations of all buildings; Calculate the average brightness of each wind field cloud image and construct a brightness loss function based on all average brightness; The true and false loss function, the hidden layer loss function, the wake vortex gradient loss function, the empirical mode feature loss function, the building position loss function and the brightness loss function are combined to obtain a comprehensive loss function.

3. The wind field prediction method according to claim 2, characterized in that: The true and false loss function is: in, Represents the true and false loss function value of the generator, Represents the true and false loss function value of the discriminator, p z (z) represents the set of all predicted wind field cloud images, G(z) represents the z-th predicted wind field cloud image, D1(G(z)) represents the output of the discriminator for the z-th predicted wind field cloud image, || ||2 represents the L2 norm, and p data (x) represents the set of all wind field cloud images, and D1(x) represents the output of the discriminator for the wind field cloud image x; The hidden layer loss function is: Among them, L layer represents the hidden layer loss function value, o i (D1(x)) represents the hidden layer output obtained by processing the wind field cloud image x through the i-th hidden layer, o i (D1(G(z))) represents the hidden layer output obtained by processing the z-th predicted wind field cloud image by the i-th hidden layer, and k represents the number of hidden layers; The wake vortex gradient loss function is: Among them, L grad represents the value of the wake vortex gradient loss function, represents the vortex structure gradient of the wind field cloud map x, Represents the vortex structure gradient of the zth predicted wind field cloud map.

4. The wind field prediction method according to claim 3, characterized in that: The empirical mode feature loss function is: Among them, L mode represents the empirical mode feature loss function value, represents the nth-order empirical mode information of the wind field cloud image x, represents the nth-order empirical modal information of the zth predicted wind field cloud map, and m represents the order of the empirical modal information.

5. The wind field prediction method according to claim 4, characterized in that: The building position loss function is: Among them, L position represents the building location loss function value, D2(G(z)) represents all building locations in the z-th predicted wind field cloud map, z * Represents the locations of all buildings in the target area corresponding to the z-th predicted wind field cloud map.

6. The wind field prediction method according to claim 5, characterized in that: The brightness loss function is: Among them, L bright represents the brightness loss function value, c(D1(x)) represents the average brightness of the wind field cloud image x, and c(D1(G(z))) represents the average brightness of the z-th predicted wind field cloud image.

7. The wind field prediction method according to claim 6, characterized in that: The comprehensive loss function is: Among them, L G Represents the comprehensive loss function value of the generator, L D Represents the comprehensive loss function value of the discriminator.

8. A wind field prediction device for a group of buildings, characterized in that: include: An acquisition module, for acquiring the locations of multiple buildings in multiple target areas and a wind field cloud map of each of the target areas; The wind field cloud map is used to describe the wind pressure value data of the target area; A construction module is used to construct a comprehensive loss function of a generative adversarial network according to the wind field cloud map of each target area; the comprehensive loss function is used to describe the accuracy of the predicted wind field cloud map of each target area generated by the generative adversarial network under the building position, brightness, and wind field wake vortex; A training module, using the comprehensive loss function to train the generative adversarial network to obtain a trained generative adversarial network; The wind field prediction module uses the trained generative adversarial network to perform wind field prediction on the area to be predicted, and obtains the final wind field prediction cloud map of the area to be predicted.

9. A terminal device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the wind field prediction method for a group of buildings as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the wind field prediction method for a group of buildings as claimed in any one of claims 1 to 7 is implemented.

Images