A two-stage rapid prediction method for urban thermal comfort based on building and greening layout

Abstract

The present invention provides a two-stage urban thermal comfort rapid prediction method based on building and greening layout, which relates to the technical field of urban thermal comfort assessment and optimization. Traditional thermal comfort assessment methods rely on high-precision solvers for fluid mechanics calculations, have long simulation cycles and high requirements for computing resources, and are difficult to adapt to the needs of high-intensity urban renewal. This scheme uses deep learning for simulation and prediction, and predicts the impact of building layout and greening layout on the urban thermal environment in stages. It starts from the macro-building layout to obtain the overall thermal environment distribution, and then combines the micro-greening configuration to refine the thermal comfort prediction, and realizes the collaborative analysis and prediction of the macro-building layout and the micro-greening layout. This collaborative optimization method ensures the complementarity and coordination between the building and greening configurations, and effectively improves the accuracy of the UTCI distribution prediction.

Description

Technical Field

[0001] The present invention belongs to the technical field of urban thermal comfort evaluation and optimization, and in particular relates to a two-stage urban thermal comfort rapid prediction method based on building and greening layout. Background Art

[0002] With the rapid development of modern society, urban thermal environment problems are becoming increasingly prominent, especially in high-intensity built environments, where the heat island effect has a serious impact on residents' lives and building energy consumption. Traditional urban thermal comfort assessment methods mainly rely on complex physical simulation and numerical simulation tools, such as ENVI-met and Fluent. Although these methods have high accuracy in simulating urban thermal environments, they are difficult to meet the needs of rapid iteration of design solutions due to their high computational complexity and long time consumption.

[0003] There are two main problems with existing thermal comfort assessment methods: First, the computational cost is high, and it is difficult to respond quickly to design changes, especially in the process of urban renewal and transformation, there is a lack of efficient tools to support real-time optimization of solutions. Second, the data requirements are high. Existing models require a large amount of meteorological, architectural and geographical data as input, and these data are not available in all urban areas, which limits the application of the model in practice. In addition, it is difficult for existing methods to effectively combine the multi-scale characteristics of urban form and greening layout, so it is impossible to comprehensively evaluate the comprehensive impact of building layout and landscape greening on the urban thermal environment. Therefore, how to achieve rapid assessment of urban thermal comfort while ensuring prediction accuracy, especially in the process of urban built environment renewal, to perform efficient thermal environment optimization prediction for different design schemes, has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of the shortcomings of the existing technology, the present invention proposes a two-stage urban thermal comfort rapid prediction method based on building and greening layout to solve the problems of long simulation time and difficulty in supporting rapid iteration of the built environment in the urban thermal comfort assessment process. The present invention uses deep learning technology to achieve accurate prediction of thermal comfort of the urban built environment, providing scientific decision-making support for the updating and optimization of the built environment.

[0005] The two-stage rapid prediction method of urban thermal comfort based on building and greening layout includes the following steps:

[0006] S1: Constructing a building layout training dataset.

[0007] S1.1: Obtain the building layout features of several study areas, and establish a building layout map based on the building layout features; the building layout features include building height data and building outlines; the building layout map is used to represent the distribution of buildings in the study area, and map the height of the building through the depth of color.

[0008] S1.2: Obtain historical meteorological data for each study area; the historical meteorological data include air temperature, relative humidity, wind speed and mean radiation temperature.

[0009] S1.3: Perform UTCI simulation on the building layout diagram according to the acquired historical meteorological data to obtain UTCI data represented in matrix form.

[0010] S1.4: Convert the UTCI data into a UTCI distribution map of set pixels, obtain a simulated building layout UTCI distribution map and compare it with the building layout Figure 1 One to one correspondence.

[0011] S1.5: The building layout maps of different study areas and their corresponding simulated building layout UTCI distribution maps are taken as a group of samples to construct a building layout training dataset including several groups of samples.

[0012] S2: Construct greening layout training dataset.

[0013] S2.1: Obtain the greening layout characteristics of several study areas that are the same as S1.1, and establish a greening layout map based on the greening layout characteristics and building layout characteristics.

[0014] The greening layout features include the location, quantity, size and height of trees; the greening layout map is used to represent the distribution of trees and buildings in the study area, in which the trees are simplified into rectangles, and the heights of trees and buildings are mapped through the depth of color.

[0015] S2.2: Perform UTCI simulation on the greening layout map based on historical meteorological data to obtain UTCI data represented in matrix form.

[0016] S2.3: Convert the UTCI data into a UTCI distribution map of set pixels, obtain a simulated greening layout UTCI distribution map and compare it with the greening layout Figure 1 One to one correspondence.

[0017] S2.4: The greening layout map and the corresponding simulated greening layout UTCI distribution map are taken as a group of samples to construct a greening layout training data set including several groups of samples.

[0018] S3: Construct a thermal comfort prediction model.

[0019] The thermal comfort prediction model includes two branches, namely a thermal comfort prediction branch based on building layout and a thermal comfort prediction branch based on greening layout.

[0020] The thermal comfort prediction branch based on building layout is Pix2pix generative adversarial network, and the input is the building layout map The output is the predicted building layout UTCI distribution map generated based on the building layout map .

[0021] The thermal comfort prediction branch based on greening layout is an improved Pix2pix generative adversarial network, and the input is the predicted building layout UTCI distribution map generated by the thermal comfort prediction branch based on building layout Greening layout diagram ,By integrating the building layout characteristics and greening layout characteristics, a predicted greening layout UTCI distribution map is generated.

[0022] The improved Pix2pix generative adversarial network includes two parts: a generator G and a discriminator D.

[0023] The generator G adopts an improved U-Net, which includes two processing processes. The first processing process performs N downsampling processes on the input building layout UTCI distribution map, and generates building layout UTCI features of different resolutions after each downsampling process; the second processing process performs a preliminary downsampling process on the greening layout map, and fuses the greening layout features obtained after the downsampling process with the building layout UTCI features obtained after one downsampling process in the first processing process through an attention mechanism to obtain a fused feature; the fused feature continues to be downsampled multiple times, and the fused features after each downsampling process are fused with the building layout UTCI features obtained after the corresponding number of downsampling processes to obtain the final fused feature; finally, the final fused feature generates a predicted greening layout UTCI distribution map through several upsampling processes. y′ , wherein a skip connection is used between downsampling and upsampling in the second processing process; the downsampling is implemented by sequentially connected convolutional layers, batch normalization layers, and ReLU activation layers; the upsampling is implemented by sequentially connected deconvolutional layers, convolutional layers, batch normalization layers, and ReLU activation layers.

[0024] The discriminator adopts an improved PatchGAN structure to evaluate the authenticity of the predicted greening layout UTCI distribution map. Specifically, the predicted building layout UTCI distribution map Perform several downsampling to obtain the intermediate feature map; convert the predicted greening layout UTCI distribution map y′ And simulated green layout UTCI distribution map After superposition and several times of downsampling, it is fused with the intermediate feature map by addition to obtain a fused UTCI distribution map. Finally, the fused UTCI distribution map is segmented and the authenticity of each area is judged to evaluate the authenticity of the predicted greening layout UTCI distribution map. The downsampling is achieved by sequentially connected convolutional layers, batch normalization layers and ReLU activation layers.

[0025] The method for performing feature fusion through the attention mechanism is:

[0026] First calculate the attention weight matrix:

[0027] (3)

[0028] in, represents the attention weight matrix, and They represent the query matrix and key matrix from the UTCI features of the building layout, respectively.

[0029] Then the attention weight matrix is ​​used to weight the building layout UTCI features to obtain the weighted building layout UTCI features:

[0030] (4)

[0031] in, is the weighted UTCI feature of the building layout, UTCI features for building layout.

[0032] Then, the greening layout features are added element by element. UTCI characteristics of building layout after weighting Fusion is performed to obtain fusion features :

[0033] (5)

[0034] S4: Train the thermal comfort prediction model using the building layout training data set and the greening layout training data set to obtain a trained thermal comfort prediction model.

[0035] S5: Obtain the building layout characteristics and greening layout characteristics, establish the building layout map and the greening layout map and input them into the thermal comfort prediction model for prediction, obtain the predicted greening layout UTCI distribution map, and judge the thermal comfort according to the predicted greening layout UTCI distribution map.

[0036] S5.1: Establish building layout diagram and greening layout diagram based on the building layout characteristics and greening layout characteristics.

[0037] S5.2: Input the building layout diagram into the thermal comfort prediction branch based on the building layout to obtain the predicted building layout UTCI distribution diagram.

[0038] S5.3: Input the greening layout map and the predicted building layout UTCI distribution map into the thermal comfort prediction branch based on the greening layout, obtain the predicted greening layout UTCI distribution map, and judge the thermal comfort according to the predicted greening layout UTCI distribution map.

[0039] Compared with the prior art, the beneficial effects of the present invention are:

[0040] (1) Realizing macro- and micro-coordinated optimization: The present invention predicts the impact of building layout and greening layout on the urban thermal environment in stages. It starts from the macro-building layout to obtain the overall thermal environment distribution, and then combines the micro-greening configuration to refine the thermal comfort prediction, thereby realizing the coordinated analysis and prediction of the macro-building layout and the micro-greening layout. This collaborative optimization method ensures the complementarity and coordination between the building and greening configurations, and effectively improves the accuracy of UTCI distribution prediction.

[0041] (2) Significantly improve prediction efficiency: Traditional thermal comfort assessment methods rely on high-precision solvers for fluid mechanics calculations, have long simulation cycles and high requirements for computing resources, and are difficult to adapt to the needs of high-intensity urban renewal. Compared with the traditional method of using CFD tools to simulate urban thermal environments, this solution uses deep learning for simulation and prediction, which significantly shortens the calculation time. It uses a thermal comfort prediction model to directly predict thermal comfort indicators, while maintaining high accuracy. It achieves hundreds of times of efficiency improvement, adapting to the needs of rapid iteration in high-intensity urban built environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 The figure is a flow chart of a two-stage urban thermal comfort rapid prediction method based on building and greening layout in an embodiment of the present invention.

[0043] Figure 2 Schematic diagram of the structure of the thermal comfort prediction model in an embodiment of the present invention.

[0044] Figure 3 This is a structural diagram of the generator in the improved Pix2pix generative adversarial network in an embodiment of the present invention.

[0045] Figure 4 2 is a structural diagram of the discriminator in the improved Pix2pix generative adversarial network in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The present invention is described in detail below with reference to the accompanying drawings and embodiments.

[0047] A two-stage rapid prediction method for urban thermal comfort based on building and greening layout, such as Figure 1 As shown, the following steps are included:

[0048] S1: Constructing a building layout training dataset.

[0049] S1.1: Obtain the building layout characteristics of several study areas, and establish a building layout map based on the building layout characteristics; the building layout characteristics include building height data and building outlines; the building layout map is used to represent the distribution of buildings in the study area, and map the height of the building through the depth of color, reflecting the height change characteristics, which is convenient for subsequent thermal comfort prediction and analysis.

[0050] In this implementation, the building layout diagram is created by Rhino and has a size of 300m×300m.

[0051] S1.2: Obtain historical meteorological data for each study area; the historical meteorological data include air temperature, relative humidity, wind speed and mean radiation temperature.

[0052] In this implementation, historical meteorological data are mainly provided by local online meteorological databases and exported in “.epw” format to ensure that meteorological conditions can reflect the actual situation in the study area.

[0053] S1.3: Perform UTCI (Universal Thermal Climate Index) simulation on the building layout according to the acquired historical meteorological data to obtain UTCI data represented in matrix form.

[0054] The UTCI data is calculated using Honeybee, and the calculation formula is:

[0055] (1)

[0056] In the formula, is a general heat climate index, is the air temperature, is the mean radiant temperature, is the wind speed, is the relative humidity, It is a nonlinear model that integrates air temperature, mean radiant temperature, wind speed and relative humidity, and is used to evaluate the impact of different climatic conditions on human thermal comfort.

[0057] In this implementation, the Honeybee plug-in in the Grasshopper parametric platform of Rhino is used to perform UTCI simulation. The grid size of the simulation calculation is set to 5 meters × 5 meters, and the monitoring height is 1.8 meters to capture the urban microclimate changes in detail. All input parameters are set according to the Tianjin annual conventional climate data standards to ensure the accuracy of the simulation results.

[0058] S1.4: In order to match the input requirements of the deep learning model, the UTCI data is converted into a UTCI distribution map of set pixels, and the simulated building layout UTCI distribution map is obtained and compared with the building layout Figure 1 One to one correspondence.

[0059] In this embodiment, the pixel value is set to 256×256.

[0060] S1.5: The building layout maps of different study areas and their corresponding simulated building layout UTCI distribution maps are taken as a group of samples to construct a building layout training dataset including several groups of samples.

[0061] The training data set in this embodiment includes building layout maps of 1,000 random areas and their corresponding building layout UTCI distribution maps. The data set has rich building types and diverse layout features, providing sufficient support for subsequent model training.

[0062] S2: Construct greening layout training dataset.

[0063] S2.1: Obtain the greening layout characteristics of several study areas that are the same as S1.1, and establish a greening layout map based on the greening layout characteristics and building layout characteristics.

[0064] The greening layout characteristics include the location, quantity, size and height of trees, which are crucial for analyzing the impact of greening on the thermal environment; the greening layout map is used to represent the distribution of trees and buildings in the study area, in which the trees are simplified into rectangles, and the heights of trees and buildings are mapped through the depth of color.

[0065] In this embodiment, the height range of the trees is set between 3 and 10 meters. To simplify the calculation, a simplified model will be used to replace the crown of the real tree. The maximum size of the simplified model is 7 meters × 7 meters × 3 meters, and the minimum size is 3 meters × 3 meters × 1.5 meters.

[0066] S2.2: Perform UTCI simulation on the greening layout map based on historical meteorological data to obtain UTCI data represented in matrix form.

[0067] In this implementation, the UTCI simulation uses the Honeybee plug-in in the Grasshopper parametric platform of Rhino to perform UTCI simulation on fixed building layouts and different greening schemes. The grid size and parameter settings of the simulation calculation are consistent with those in S1 to ensure the consistency and reliability of the model input.

[0068] S2.3: Convert the UTCI data into a UTCI distribution map of set pixels, obtain a simulated greening layout UTCI distribution map and compare it with the greening layout Figure 1 One-to-one correspondence ensures that the input of the deep learning model meets the requirements.

[0069] In this embodiment, the pixel value is set to 256×256.

[0070] S2.4: The greening layout map and the corresponding simulated greening layout UTCI distribution map are taken as a group of samples to construct a greening layout training data set including several groups of samples.

[0071] In this embodiment, the greening layout training data set includes 1,000 greening layout maps with trees of different numbers, sizes and positions and their corresponding greening layout UTCI distribution maps.

[0072] S3: Construct a thermal comfort prediction model.

[0073] like Figure 2 As shown, the thermal comfort prediction model includes two branches, namely a thermal comfort prediction branch based on building layout and a thermal comfort prediction branch based on greening layout.

[0074] The thermal comfort prediction branch based on building layout is Pix2pix generative adversarial network (GAN), and the input is the building layout map The output is the predicted building layout UTCI distribution map generated based on the building layout map .

[0075] The thermal comfort prediction branch based on greening layout is an improved Pix2pix generative adversarial network, and the input is the predicted building layout UTCI distribution map generated by the thermal comfort prediction branch based on building layout Greening layout diagram ,By integrating the building layout characteristics and greening layout characteristics, a predicted greening layout UTCI distribution map is generated.

[0076] The improved Pix2pix generative adversarial network includes two parts: a generator G and a discriminator D.

[0077] like Figure 3As shown, the generator G adopts an improved U-Net, which includes two processing processes. The first processing process performs N downsampling processing on the input building layout UTCI distribution map, and generates building layout UTCI features of different resolutions after each downsampling process; the second processing process performs a preliminary downsampling process on the greening layout map, and fuses the greening layout features obtained after the downsampling process with the building layout UTCI features obtained after one downsampling process in the first processing process through the attention mechanism to obtain a fused feature; the fused feature continues to be downsampled multiple times, and the fused features after each downsampling process are fused with the building layout UTCI features obtained after the corresponding number of downsampling processes to ensure that the influence of the building layout on the greening layout is fully reflected in the whole process, and the final fused feature is obtained; finally, the final fused feature is upsampled several times to generate a predicted greening layout UTCI distribution map y′ , where Skip Connections are used between downsampling and upsampling in the second processing process.

[0078] The downsampling is achieved by sequentially connected convolution layers, batch normalization layers, and ReLU activation layers; the upsampling is achieved by sequentially connected deconvolution layers, convolution layers, batch normalization layers, and ReLU activation layers to gradually restore the spatial dimensions and detail information of the features; the purpose of this is to extract features through multi-layer convolution operations, retain and transfer features at different resolutions to the upsampling stage through jump connections, and at the same time combine the attention mechanism to strengthen the feature extraction of the predicted building layout UTCI distribution map generated by the thermal comfort prediction branch based on the building layout.

[0079] The generation process of the generator is expressed as:

[0080] (2)

[0081] in, represents the generator, which is based on the improved U-Net structure, represents the predicted building layout UTCI distribution map; It represents the greening layout diagram; UTCI distribution map showing the predicted greening layout.

[0082] The method for performing feature fusion through the attention mechanism is:

[0083] First calculate the attention weight matrix:

[0084] (3)

[0085] in, represents the attention weight matrix, and They represent the query matrix and key matrix from the building layout UTCI features respectively; by calculating the correlation between the building layout UTCI features and normalizing them through SoftMax, each attention score is converted into a probability value, thereby obtaining an attention weight matrix for weighting the building layout UTCI features, ensuring that important features are given priority in subsequent processing.

[0086] Then, the attention weight matrix is ​​used to weight the building layout UTCI features to obtain the weighted building layout UTCI features;

[0087] (4).

[0088] in, is the weighted UTCI feature of the building layout, UTCI features for building layout.

[0089] Then, the greening layout features are added element by element. UTCI characteristics of building layout after weighting Fusion is performed to obtain fusion features ;

[0090] (5)

[0091] like Figure 4 As shown in the figure, the discriminator adopts an improved PatchGAN structure to evaluate the authenticity of the predicted greening layout UTCI distribution map. Specifically, conditional information (predicted building layout UTCI distribution map) is introduced to predict the building layout UTCI distribution map. Perform several downsampling to obtain the intermediate feature map; convert the predicted greening layout UTCI distribution map y′ And simulated green layout UTCI distribution map After superposition and several times of downsampling, it is fused with the intermediate feature map by addition to obtain the fused UTCI distribution map. Finally, the fused UTCI distribution map is segmented and the authenticity of each area (Patch) is judged to realize the authenticity evaluation of the predicted greening layout UTCI distribution map.

[0092] The downsampling is achieved by sequentially connecting convolutional layers, batch normalization layers and ReLU activation layers.

[0093] The input of the discriminator is expressed as:

[0094] (6)

[0095] in, represents the combined input for the joint evaluation of the predicted greening layout UTCI distribution map and the actual greening layout UTCI distribution map, i.e., the input of the discriminator; y It represents the real greening layout UTCI distribution map, that is, the simulated greening layout UTCI distribution map.

[0096] S4: Train the thermal comfort prediction model using the building layout training data set and the greening layout training data set to obtain a trained thermal comfort prediction model.

[0097] In the training process of the thermal comfort prediction branch based on building layout, the generator of the Pix2pix generative adversarial network is responsible for converting the input building layout map into the predicted building layout UTCI distribution map. The generator uses six layers of jump connections to retain the detailed information of the building layout features to ensure that important features are transmitted at different resolutions and used for the final output. At the same time, the discriminator uses the simulated building layout UTCI distribution map in the building layout training dataset to evaluate the authenticity of the generated predicted building layout UTCI distribution map, so as to train the discriminant ability of the discriminator, optimize the output effect of the generator, ensure that the generated image is as close as possible to the real building layout thermal environment, and provide macro guidance for the subsequent greening layout model.

[0098] In the training process of the thermal comfort prediction branch based on greening layout, the UTCI distribution map of the building layout predicted by the thermal comfort prediction branch based on building layout ( ) and greening layout diagram ( x 2 ) as multi-source input to generate the predicted greening layout UTCI distribution map under the corresponding greening layout map ( ), the generator adopts six-layer skip connection to better combine the information of macro building layout and micro greening configuration to ensure the accuracy of the prediction results. The building layout UTCI distribution map (macro UTCI distribution map) predicted by the thermal comfort prediction branch based on building layout is used as the macro guidance input of the thermal comfort prediction branch based on greening layout. The discriminator evaluates the authenticity of the generated predicted greening layout UTCI distribution map with the greening layout UTCI distribution image simulated in the greening layout training dataset to train the discriminant ability of the discriminator and optimize the output effect of the generator.

[0099] S5: Obtain the building layout characteristics and greening layout characteristics, establish the building layout map and the greening layout map and input them into the thermal comfort prediction model for prediction, obtain the predicted greening layout UTCI distribution map, and judge the thermal comfort according to the predicted greening layout UTCI distribution map.

[0100] A two-stage deep learning model is used to predict the UTCI distribution for macro-scale building layout and micro-scale greening layout, respectively.

[0101] S5.1: Establish building layout diagram and greening layout diagram based on the building layout characteristics and greening layout characteristics.

[0102] S5.2: Input the building layout diagram into the thermal comfort prediction branch based on the building layout to obtain the predicted building layout UTCI distribution diagram.

[0103] S5.3: The greening layout map and the predicted building layout UTCI distribution map are input into the thermal comfort prediction branch based on the greening layout, and the predicted building layout UTCI distribution map and the greening layout map are combined to obtain the predicted greening layout UTCI distribution map and judge the thermal comfort based on the predicted greening layout UTCI distribution map.

Claims

1. A two-stage urban thermal comfort rapid prediction method based on building and greening layout, characterized in that: The steps include: S1: construct a building layout training data set; the building layout training data set includes a number of building layout diagrams and corresponding simulated building layout UTCI distribution diagrams; S2: constructing a greening layout training data set; the greening layout training data set includes a number of greening layout diagrams and corresponding simulated greening layout UTCI distribution diagrams; S3: Construct thermal comfort prediction model; The thermal comfort prediction model includes two branches, namely a thermal comfort prediction branch based on building layout and a thermal comfort prediction branch based on greening layout; The thermal comfort prediction branch based on building layout is used to predict UTCI according to the input building layout diagram to obtain a predicted building layout UTCI distribution diagram; The thermal comfort prediction branch based on greening layout is used to generate a predicted building layout UTCI distribution map and a greening layout map according to the predicted building layout UTCI distribution map and the greening layout map generated by the thermal comfort prediction branch based on the building layout, by fusing the building layout characteristics and the greening layout characteristics to generate a predicted greening layout UTCI distribution map; S4: training the thermal comfort prediction model using the building layout training data set and the greening layout training data set to obtain a trained thermal comfort prediction model; S5: Obtain the building layout characteristics and greening layout characteristics, establish the building layout map and the greening layout map and input them into the thermal comfort prediction model for prediction, obtain the predicted greening layout UTCI distribution map, and judge the thermal comfort according to the predicted greening layout UTCI distribution map.

2. The two-stage urban thermal comfort rapid prediction method based on building and greening layout according to claim 1 is characterized in that: S1 specifically includes: S1.1: Obtaining building layout features of several study areas, and establishing a building layout map according to the building layout features; the building layout features include building height data and building outlines; the building layout map is used to represent the distribution of buildings in the study area, and maps the height of the buildings through the depth of color; S1.2: Obtain historical meteorological data for each study area; the historical meteorological data include air temperature, relative humidity, wind speed and mean radiation temperature; S1.3: Perform UTCI simulation on the building layout diagram according to the acquired historical meteorological data to obtain UTCI data represented in matrix form; S1.4: converting the UTCI data into a UTCI distribution map of set pixels, obtaining a simulated building layout UTCI distribution map and making a one-to-one correspondence with the building layout map; S1.5: The building layout maps of different study areas and their corresponding simulated building layout UTCI distribution maps are taken as a group of samples to construct a building layout training dataset including several groups of samples.

3. The two-stage urban thermal comfort rapid prediction method based on building and greening layout according to claim 2 is characterized in that: S2 specifically includes: S2.1: Obtain the greening layout characteristics of several study areas that are the same as S1.1, and establish a greening layout map based on the greening layout characteristics and building layout characteristics; The greening layout features include the location, quantity, size and height of trees; the greening layout map is used to represent the distribution of trees and buildings in the study area, in which trees are simplified into rectangles, and the heights of trees and buildings are mapped by the depth of color; S2.2: Perform UTCI simulation on the greening layout map based on historical meteorological data to obtain UTCI data represented in matrix form; S2.3: converting the UTCI data into a UTCI distribution map of set pixels, obtaining a simulated greening layout UTCI distribution map and making a one-to-one correspondence with the greening layout map; S2.4: The greening layout map and the corresponding simulated greening layout UTCI distribution map are taken as a group of samples to construct a greening layout training data set including several groups of samples.

4. The two-stage urban thermal comfort rapid prediction method based on building and greening layout according to claim 1 is characterized in that: The thermal comfort prediction branch based on building layout is a Pix2pix generative adversarial network.

5. The two-stage urban thermal comfort rapid prediction method based on building and greening layout according to claim 1 is characterized in that: The thermal comfort prediction branch based on greening layout is an improved Pix2pix generative adversarial network, which includes a generator G and a discriminator D.

6. The two-stage urban thermal comfort rapid prediction method based on building and greening layout according to claim 5 is characterized in that: The generator G adopts an improved U-Net, which includes two processing processes. The first processing process performs N downsampling processes on the input building layout UTCI distribution map, and generates building layout UTCI features of different resolutions after each downsampling process; In the second processing process, the greening layout map is subjected to a preliminary downsampling process, and the greening layout features obtained after the downsampling process are fused with the building layout UTCI features obtained after the downsampling process in the first processing process through the attention mechanism to obtain the fused features. The fused features are further downsampled multiple times, and the fused features after each downsampling process are fused with the building layout UTCI features obtained after the corresponding number of downsampling processes to obtain the final fused features; finally, the final fused features are upsampled several times to generate the predicted green layout UTCI distribution map y′ , wherein a skip connection is used between downsampling and upsampling in the second processing process; the downsampling is implemented by sequentially connected convolutional layers, batch normalization layers, and ReLU activation layers; the upsampling is implemented by sequentially connected deconvolutional layers, convolutional layers, batch normalization layers, and ReLU activation layers.

7. The two-stage urban thermal comfort rapid prediction method based on building and greening layout according to claim 5 is characterized in that: The discriminator adopts an improved PatchGAN structure to evaluate the authenticity of the predicted greening layout UTCI distribution map. Specifically, the predicted building layout UTCI distribution map Perform several downsampling to obtain the intermediate feature map; convert the predicted greening layout UTCI distribution map y′ And simulated green layout UTCI distribution map After superposition and several times of downsampling, it is fused with the intermediate feature map by addition to obtain a fused UTCI distribution map. Finally, the fused UTCI distribution map is segmented and the authenticity of each area is judged to evaluate the authenticity of the predicted greening layout UTCI distribution map. The downsampling is achieved by sequentially connected convolutional layers, batch normalization layers and ReLU activation layers.

8. The two-stage urban thermal comfort rapid prediction method based on building and greening layout according to claim 6 is characterized in that: The method for performing feature fusion through the attention mechanism is: First calculate the attention weight matrix: （3） in, represents the attention weight matrix, and They represent the query matrix and key matrix derived from the UTCI features of the building layout, respectively; Then, the attention weight matrix is ​​used to weight the building layout UTCI features to obtain the weighted building layout UTCI features; （4） in, is the weighted UTCI feature of the building layout, Lay out UTCI features for the building; Then, the greening layout features are added element by element. UTCI characteristics of building layout after weighting Fusion is performed to obtain fusion features ; （5）。 9. The two-stage urban thermal comfort rapid prediction method based on building and greening layout according to claim 1 is characterized in that: S5 specifically includes: S5.1: Establish building layout diagram and greening layout diagram according to building layout characteristics and greening layout characteristics; S5.2: Input the building layout diagram into the thermal comfort prediction branch based on the building layout to obtain the predicted building layout UTCI distribution diagram; S5.3: Input the greening layout map and the predicted building layout UTCI distribution map into the thermal comfort prediction branch based on the greening layout, obtain the predicted greening layout UTCI distribution map and judge the thermal comfort according to the predicted greening layout UTCI distribution map.

Images