Urban design parameter optimization method and system based on high-resolution thermal environment rapid prediction

Abstract

The invention discloses an urban design parameter optimization method and system based on high-resolution thermal environment rapid prediction, and the method comprises the steps: designing a plurality of simulation scenes according to the infrastructure form of a target region, carrying out the CFD simulation, constructing an original high-dimensional data set matrix, and then projecting the original high-dimensional data set matrix to a low-dimensional space, obtaining a dimensionality-reduced low-dimensional feature data set matrix, using the dimensionality-reduced low-dimensional feature data set matrix for prediction training of a radial basis function neural network model, and predicting a new design scene of a target area by the trained model to obtain a predicted value of a low-dimensional new principal component score; and restoring a complete high-resolution temperature field and wind speed field spatial distribution diagram in a new design scene through dimension raising reconstruction, and finally constructing a comprehensive evaluation index system by combining the average temperature, the wind speed and the construction and maintenance cost, thereby recommending an optimal city design parameter scheme. The urban thermal environment prediction efficiency and precision are improved, and the urban design scheme can be optimized according to the prediction result.

Description

Technical Field

[0001] This invention belongs to the field of smart city design and computer simulation technology, specifically relating to a method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment. Background Technology

[0002] With the acceleration of urbanization, the urban heat island effect is becoming increasingly prominent, posing a serious threat to residents' health and energy consumption. Obtaining high-resolution spatiotemporal distribution of the thermal environment is a prerequisite for precise regulation. Currently, obtaining high-resolution thermal environment distribution mainly relies on computational fluid dynamics simulation methods, but these methods suffer from inherent technical bottlenecks such as high computational costs and enormous time consumption, severely limiting their application in rapid scheme comparison and optimization during the urban planning and design phase.

[0003] In recent years, artificial intelligence technology has developed rapidly, offering a potential solution to the efficiency bottleneck of computational fluid dynamics simulation methods. While existing technologies attempt to apply machine learning to predict urban thermal environments, they still face two major technical drawbacks: 1) predictions based on single-point monitoring data cannot generate high-resolution spatial fields; 2) machine learning models that directly process high-dimensional spatial fields face the curse of high-dimensional data, resulting in high training costs, complex and inefficient models, and difficulty in guaranteeing the global spatial accuracy of prediction results.

[0004] Therefore, there is an urgent need in this field for an innovative technical solution that can balance prediction efficiency, accuracy, and spatial integrity. Summary of the Invention

[0005] Based on this, the present invention aims to provide a method and system for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment. By integrating computational fluid dynamics and artificial intelligence technologies, it can improve the efficiency and accuracy of urban thermal environment prediction and optimize urban design schemes based on the prediction results.

[0006] To solve the above-mentioned technical problems and achieve the above-mentioned technical effects, the present invention is implemented through the following technical solution: A method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment includes the following steps: S100. Different urban design parameters and their variables are proposed for the infrastructure form in the target area, and several simulation scenarios are designed accordingly. Each simulation scenario has scenario parameters containing its own corresponding urban design parameters and their variable values. S200. Perform CFD simulations on all the simulation scenarios in sequence to obtain the CFD simulation data of each grid node in all the simulation scenarios, and use this to construct the original high-dimensional dataset matrix, that is, obtain the high-resolution thermal environment database. S300. By performing principal component analysis (PCA) on the high-resolution thermal environment database, the original high-dimensional dataset matrix is ​​projected into a low-dimensional space to obtain a dimensionality-reduced low-dimensional feature dataset matrix. S400. The radial basis function neural network model is trained using the low-dimensional feature dataset matrix, so that the model learns the nonlinear mapping relationship from the urban design parameter vector to the principal component score vector, until the model can accurately output the predicted value of the low-dimensional principal component score according to the urban design parameters. S500: The trained radial basis function neural network model is used to predict new design scenarios in the target area where the design parameters remain unchanged and only the variables change, thereby obtaining the predicted value of the low-dimensional new principal component score of the target area under the new urban design parameters. S600. Using the PCA reconstruction algorithm, the predicted values ​​of the low-dimensional new principal component score vector are reconstructed to restore the complete high-resolution temperature field and wind speed field spatial distribution map of the target area under the new design scenario. S700. Based on the high-resolution temperature and wind speed field spatial distribution maps, combined with average temperature, wind speed, and construction and maintenance costs, a comprehensive evaluation index system is constructed, and the optimal urban design parameter scheme is recommended accordingly.

[0007] Furthermore, in step S100, the method for designing the simulated scenario is specifically as follows: First, based on the infrastructure morphology within the target area, several optimizable urban design parameters are proposed, including buildings, roads, waterways, and green spaces. Then, design variables based on the proposed urban design parameters, and select several variable values ​​for each of the proposed urban design parameters. Finally, m simulation scenarios are formed according to the full factorial design method. The number of simulation scenarios m is equal to the product of the number of variables of each of the urban design parameter items, and m is a positive integer. The variable values ​​of all the city design parameters in each simulation scenario together constitute the scenario parameters of that simulation scenario.

[0008] Furthermore, the method for establishing the variables of the urban design parameter items is as follows: When a building within the target area is selected as one of the urban design parameters, the building height and / or building density are first set as variables, and then several different building height values ​​and / or different building density values ​​are selected as the variable values ​​of the urban design parameter. When a road within the target area is selected as one of the urban design parameters, the road width and / or road area are first set as variables, and then several different road width values ​​and / or different road density values ​​are selected as the variable values ​​of the urban design parameter. When the water area within the target area is selected as an urban design parameter, the water area and / or water density are first set as variables, and then several different water area values ​​and / or different water density values ​​are selected as variable values ​​for that urban design parameter. When greening within the target area is selected as an urban design parameter, the location of the greening and / or the green coverage rate are first set as variables, and then several different greening locations and / or different green coverage rate values ​​are selected as variable values ​​for that urban design parameter.

[0009] Furthermore, in step S200, the method for constructing the high-resolution thermal environment database is specifically as follows: First, computational fluid dynamics (CFD) is used to perform geometric modeling, mesh generation, and thermal environment simulation on all m simulation scenarios in sequence. By generating multiple high-resolution thermal environment scenarios covering different scenario parameters, CFD simulation data of all N mesh nodes in each simulation scenario is obtained and denoted as mesh node data, where m and N are both positive integers. Then, all the grid node data are organized, and the grid node data belonging to the same simulation scenario are combined to form m data matrices with a dimension of 5×N, which are denoted as high-resolution thermal environment data matrices. Each simulation scenario corresponds to one high-resolution thermal environment data matrix. Finally, all m high-resolution thermal environment data matrices are combined to form an original high-dimensional dataset matrix X∈R^(m×5×N), which is the high-resolution thermal environment database.

[0010] Furthermore, each of the grid node data contains 5 characteristic parameters, specifically including spatial coordinate data (x, y, z) used to describe the geometric position of the grid node at a plane 1.5 meters above the ground, and air temperature data T and wind speed data U used to describe the thermal environment characteristics of the grid node at a plane 1.5 meters above the ground. Therefore, the grid node data is denoted as [x, y, z, T, U].

[0011] Furthermore, in step S300, the specific steps of the principal component analysis are as follows: S301. The original high-dimensional dataset matrix is ​​preprocessed using the Z-score method to obtain a standardized data matrix; S302. Based on the obtained standardized data matrix, perform covariance calculation to obtain the covariance matrix of the standardized data matrix; S303. Based on the obtained covariance matrix, perform eigenvalue decomposition to obtain several eigenvalues ​​and eigenvectors. Then, arrange the eigenvalues ​​in descending order and define the corresponding eigenvectors as principal components. S304. First, calculate the variance contribution rate of each principal component. Then, using the obtained variance contribution rate of each principal component, select the first k principal components in sequence to calculate the cumulative explained variance contribution rate, where k is a positive integer. Finally, when the calculated cumulative explained variance contribution rate is ≥95%, stop the calculation and select the k principal components corresponding to the current cumulative explained variance contribution rate. S305. Construct a projection matrix belonging to each of the simulated scenarios using the selected first k principal components, and then combine the projection matrices of each of the simulated scenarios to finally obtain the low-dimensional feature dataset matrix of all the simulated scenarios.

[0012] Furthermore, in step S400, the architecture of the radial basis function neural network model includes: The input layer is responsible for receiving new variable values ​​of the set design parameters during prediction, thereby forming a new urban design parameter vector, which serves as the input to the radial basis function neural network model. During training, a number of sample data are selected from the standardized data matrix using the K-means clustering algorithm, thereby forming a training urban design parameter vector, which also serves as the input to the radial basis function neural network model. The sample data are the standardized grid node data and their scene parameters at K different locations in the standardized data matrix. The hidden layer, containing several nodes, is responsible for mapping the urban design parameter vector input from the input layer to a low-dimensional space through nonlinear transformation, thereby achieving data dimensionality reduction and obtaining the predicted value of the principal component score vector. The output layer is responsible for outputting the predicted values ​​of the principal component score vectors obtained from the hidden layer, forming the principal component score vectors as the output of the radial basis function neural network model.

[0013] Furthermore, in step S400, the training method for the radial basis function neural network model is as follows: First, the input layer uses the K-means clustering algorithm to select new grid node data and their scene parameters at K different locations within the target area from the standardized data matrix as new sample data, thereby forming a training urban design parameter vector as input; Then, the radial basis function neural network model is trained using the low-dimensional feature dataset matrix, so that the hidden layer learns the nonlinear mapping relationship from the urban design parameter vector to the principal component score vector; Finally, the radial basis function neural network model can accurately output the predicted values ​​of low-dimensional principal component scores based on the urban design parameters, and the output layer is used to generate k principal component score vectors.

[0014] Furthermore, in step S700, the method for recommending the optimal urban design parameter scheme is specifically as follows: First, based on the complete high-resolution temperature and wind speed field spatial distribution map of the target area under the new design scenario obtained by reconstruction, calculate the average temperature and average wind speed of the target area under the new design scenario; Then, the average temperature and average wind speed of the target area under the new design scenario are combined with the construction and maintenance costs to construct a comprehensive evaluation index system; Next, the entropy weight method is used to determine the objective weight of each indicator in the comprehensive evaluation index system; Finally, the scores of the comprehensive evaluation index system for all different design scenarios are calculated, and the optimal urban design parameter scheme is recommended based on the scores.

[0015] A system for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment includes a high-resolution thermal environment database module, a high-dimensional data dimensionality reduction and feature extraction module, a low-dimensional feature prediction module, a high-resolution thermal environment reconstruction module, and a multi-objective optimization decision module; wherein, The high-resolution thermal environment database module is responsible for proposing different urban design parameters and variables for the infrastructure form in the target area, and designing several simulation scenarios based on these parameters. Then, CFD simulations are performed on all the simulation scenarios in sequence to obtain the CFD simulation data of each grid node in all the simulation scenarios. Based on this, the original high-dimensional dataset matrix is ​​constructed, which is the high-resolution thermal environment database. The high-dimensional data dimensionality reduction and feature extraction module is responsible for performing principal component analysis (PCA) on the high-resolution thermal environment database, projecting the original high-dimensional dataset matrix into a low-dimensional space to obtain a dimensionality-reduced low-dimensional feature dataset matrix. The low-dimensional feature prediction module is responsible for training the radial basis function neural network model using the low-dimensional feature dataset matrix, enabling the model to learn the nonlinear mapping relationship from the urban design parameter vector to the principal component score vector, until the model can accurately output the predicted value of the low-dimensional principal component score based on the urban design parameters. On the other hand, it is responsible for using the trained radial basis function neural network model to predict the new design scenario of the target area, and obtaining the predicted value of the low-dimensional new principal component score of the target area under the new urban design parameters. The high-resolution thermal environment reconstruction module is responsible for using the PCA reconstruction algorithm to reconstruct the predicted values ​​of the low-dimensional new principal component score vectors in a higher dimension, thereby restoring the complete high-resolution temperature field and wind speed field spatial distribution map of the target area under the new design scenario. The multi-objective optimization decision-making module is responsible for constructing a comprehensive evaluation index system based on the high-resolution temperature and wind speed field spatial distribution maps, combined with average temperature, wind speed, and construction and maintenance costs, and recommending the optimal urban design parameter scheme accordingly.

[0016] A computer device includes a processor, a memory, a communication interface, and a communication bus. The processor, the memory, and the communication interface communicate with each other via the communication bus. The memory stores at least one executable instruction, which causes the processor to perform the operation corresponding to the above-mentioned urban design parameter optimization method based on rapid prediction of high-resolution thermal environment.

[0017] A computer-readable storage medium storing at least one executable instruction that causes a processor to perform the operations corresponding to the above-described method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment.

[0018] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Improved computational efficiency: This invention transforms the complex spatial field prediction problem into a regression problem in a low-dimensional space through a "high-dimensional-low-dimensional-high-dimensional" technical path, which greatly reduces the learning difficulty and computational burden of machine learning models.

[0019] 2. High-precision spatial fidelity: This invention does not sacrifice resolution for efficiency, but rather accurately restores a high-resolution spatial cloud map with a precision comparable to that of CFD simulation through data dimensionality reconstruction, preserving key spatial details and local features, and providing a reliable basis for refined design.

[0020] 3. System optimization decision-making capability: This invention combines rapid prediction results with multi-objective decision-making methods to achieve a closed loop from "prediction" to "optimization". It can automatically recommend design schemes that achieve the best balance between environmental benefits and economic costs, providing intuitive and quantitative decision support for urban planning.

[0021] 4. Strong versatility and scalability: The technical framework of this invention is not limited to thermal environment prediction, but can be extended to the efficient simulation and optimization of other urban physical fields such as noise simulation, pollutant diffusion, and solar radiation analysis, and has broad application prospects.

[0022] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the invention and to implement it according to the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings. Specific embodiments of the present invention are given in detail below with reference to the accompanying drawings. Attached Figure Description

[0023] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings: Figure 1 This is a flowchart of the urban design parameter optimization system and method based on high-resolution thermal environment rapid prediction, as described in this invention.

[0024] Figure 2 This is a schematic diagram illustrating the calculation process of the radial basis function neural network model in an embodiment of the present invention.

[0025] Figure 3 This is a complete high-resolution spatial distribution map of the temperature field of the community under the predicted new design scenario in this embodiment of the invention.

[0026] Figure 4 This is a graph showing the comprehensive evaluation results of all green coverage rates calculated in this embodiment of the invention. Detailed Implementation

[0027] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings to provide a clearer understanding of the invention's purpose, features, and advantages. It should be understood that the embodiments shown in the drawings are not intended to limit the scope of the invention, but are merely illustrative of the essential spirit of the invention's technical solution.

[0028] In the following description, certain specific details are set forth for the purpose of illustrating various disclosed embodiments in order to provide a thorough understanding of the various disclosed embodiments. However, those skilled in the art will recognize that embodiments may be practiced without one or more of these specific details. In other instances, well-known apparatuses, structures, and techniques associated with this application may not have been shown or described in detail to avoid unnecessarily obscuring the description of the embodiments.

[0029] Unless the context requires otherwise, throughout the specification and claims, the word “comprising” and its variations, such as “including” and “having”, shall be understood to have an open, inclusive meaning, that is, to be interpreted as “including, but not limited to”.

[0030] Throughout this specification, references to "an embodiment" or "an embodiment" indicate that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment. Therefore, the appearance of "in an embodiment" or "an embodiment" in various places throughout the specification does not necessarily refer to the same embodiment. Furthermore, a particular feature, structure, or characteristic may be combined in any manner in one or more embodiments.

[0031] The singular forms “a” and “the” used in this specification and the appended claims include plural references unless otherwise expressly stated herein. It should be noted that the term “or” is generally used to mean “and / or” unless otherwise expressly stated herein.

[0032] Furthermore, the technical features involved in the different embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0033] See Figure 1 As shown, this invention provides a system and method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment. The system mainly includes a high-resolution thermal environment database module, a high-dimensional data dimensionality reduction and feature extraction module, a low-dimensional feature prediction module, a high-resolution thermal environment reconstruction module, and a multi-objective optimization decision module.

[0034] The method of this invention is mainly implemented through the above-mentioned functional modules, and specifically includes the following steps: S100: The high-resolution thermal environment database module takes the city, district, or community to be optimized as the target area, proposes different urban design parameters and their variables for one or more infrastructures such as buildings, roads, water bodies, and green spaces (such as trees and grasslands) within the target area, and designs a certain number of simulation scenarios based on these parameters. Each simulation scenario has scenario parameters containing its corresponding urban design parameters and their variable values.

[0035] The specific design method for the simulated scenario in this step is as follows: First, n types of infrastructure within the target area, such as buildings, roads, waterways, and green infrastructure (e.g., trees, grass), are designated as optimizable urban design parameters (e.g., labeled P1, P2, ..., Pn). Then, variables are designed based on the proposed urban design parameters, and q variable values ​​are selected for each variable (e.g., q1 variable values ​​are set for urban design parameter P1, labeled P11, P12, ..., P1). q1 The urban design parameter P2 has q2 variable values, labeled as P21, P22, ..., P2 q2 The urban design parameter Pn has qn variable values, labeled as Pn1, Pn2, ..., Pn qn Finally, based on the full factorial design method, a total of m = q1 × q2 × … × qn simulation scenarios are formed. That is, the number of simulation scenarios is equal to the product of the number of variables of each city design parameter item, and the variable values ​​of all city design parameter items in each simulation scenario constitute the scenario parameters of that simulation scenario, where n, q and m are all positive integers.

[0036] The urban design parameters must be at least one of the following in the target area: buildings, roads, waterways, and green infrastructure (such as trees and grass).

[0037] For example, when selecting buildings within a target area as urban design parameters, you can first set building height and / or building density as variables, and then select several different building height values ​​and / or different building density values ​​as variable values ​​for that urban design parameter.

[0038] For example, when selecting roads within the target area as urban design parameters, you can first set the road width and / or road area as variables, and then select several different road width values ​​and / or different road density values ​​as variable values ​​for the urban design parameter.

[0039] For example, when selecting water bodies within the target area as urban design parameters, the water area and / or water density can be set as variables first, and then several different water area values ​​and / or different water density values ​​can be selected as variable values ​​for the urban design parameter.

[0040] For example, when selecting green infrastructure (such as trees, grass, etc.) in the target area as urban design parameters, the location of greenery and / or the green coverage rate can be set as variables first, and then several different green locations and / or different green coverage rate values ​​can be selected as variable values ​​for the urban design parameter.

[0041] S200: The high-resolution thermal environment database module performs CFD simulations on all simulation scenarios in sequence to obtain CFD simulation data for each grid node in all simulation scenarios. Based on this, an original high-dimensional dataset matrix containing the spatial coordinates, three-dimensional temperature field, and wind speed field data of each grid node located at the pedestrian height plane is constructed, thus obtaining the high-resolution thermal environment database.

[0042] The specific method for constructing the high-resolution thermal environment database in this step is as follows: First, computational fluid dynamics (CFD) is used to perform geometric modeling, mesh generation, and thermal environment simulation for all m simulation scenarios in sequence. By generating multiple high-resolution thermal environment scenarios covering different scenario parameters, CFD simulation data of all N mesh nodes in each simulation scenario is obtained and denoted as mesh node data, where m and N are both positive integers.

[0043] Each grid node contains five feature parameters, including spatial coordinate data (x, y, z) describing the geometric position of the grid node at a plane 1.5 meters above the ground, and air temperature data T and wind speed data U describing the thermal environment characteristics of the grid node at a plane 1.5 meters above the ground. Therefore, the grid node data is denoted as [x, y, z, T, U].

[0044] Then, all grid node data are organized, and grid node data belonging to the same simulation scenario are combined to form m data matrices with dimensions of 5×N, denoted as high-resolution thermal environment data matrices. Each simulation scenario corresponds to a high-resolution thermal environment data matrix.

[0045] Finally, all m high-resolution thermal environment data matrices are combined to form an original high-dimensional dataset matrix X∈R^(m×5×N), which is the high-resolution thermal environment database.

[0046] S300: The high-dimensional data dimensionality reduction and feature extraction module performs principal component analysis (PCA) on the original high-dimensional dataset matrix X, projecting the original high-dimensional dataset matrix to a low-dimensional space to obtain the dimensionality-reduced low-dimensional feature dataset matrix ℾ.

[0047] The specific steps of PCA analysis in this process are as follows: S301, Data Standardization; Because the units and numerical ranges of the five feature parameters in the grid node data differ, the Z-score method is used to preprocess the original high-dimensional dataset matrix X to obtain a standardized data matrix Z. This ensures that the analysis will not be biased towards features with larger numerical values ​​during data processing, thus guaranteeing the accuracy of the analysis results. The specific calculation process is as follows: (1); (2); (3); In formulas (1)~(3), Z ij This represents the j-th feature value of the i-th sample after standardization. X ij This represents the j-th feature value of the i-th sample in the original data. μ j Let the j-th feature be the mean of all samples. σ j Let represent the standard deviation of the j-th feature across all samples, where i = 1, 2, ..., p, j = 1, 2, ..., p. m This represents the total number of simulation scenarios, where i, j, p, and m are all positive integers; S302, Covariance matrix calculation; Based on the obtained standardized data matrix Z, the covariance is calculated to obtain the covariance matrix R of the standardized data matrix Z. The specific formula is as follows: (4); In formula (4), R The covariance matrix represents the normalized data matrix. Z Represents a standardized data matrix. This represents the transpose of a normalized data matrix. m This represents the total number of simulated scenarios, where m is a positive integer; S303, Characteristic Decomposition; Based on the obtained covariance matrix R, eigenvalue decomposition is performed to obtain several eigenvalues ​​(variance magnitude) and eigenvectors (principal component directions). Then, the eigenvalues ​​are arranged in descending order, and the corresponding eigenvectors are defined as principal components. The specific formula is as follows: (5); In formula (5), R The covariance matrix represents the normalized data matrix. λ j Represents the j-th eigenvalue. v j Let represent the corresponding eigenvector, where j = 1, 2, ..., p; and the eigenvalues ​​are arranged in descending order. λ 1≥ λ 2≥...≥ λ p ≥0, the eigenvectors corresponding to the eigenvalues v 1. v 2、...、 vp They are defined as principal components, where j and p are both positive integers; S304, Principal Component Selection; In order to retain as much useful information as possible while reducing data dimensionality, the most important principal components were selected based on their ability to explain variance. First, calculate the variance contribution rate of each principal component, using the following formula: (6); In formula (6), a j This represents the variance contribution rate of the j-th principal component. λ j Let j represent the j-th eigenvalue, where j = 1, 2, ..., p. k Indicates the number of all principal components. λ k Let i represent the k-th eigenvalue, where i, j, p, and k are all positive integers. Then, using the variance contribution rate of each principal component, the cumulative explained variance contribution rate is calculated for the first k principal components selected sequentially each time, where k is a positive integer. The specific formula is as follows: (7); In formula (6), β k This represents the cumulative variance contribution rate explained by the first k principal components. a j This represents the variance contribution rate of the j-th principal component. k This indicates the number of principal components selected, where j and k are both positive integers; Finally, when the calculated cumulative explained variance contribution rate is ≥95%, the calculation is stopped, and the k principal components corresponding to the current cumulative explained variance contribution rate are selected. S305, dimension reduction projection; Construct the projection matrix W = [ using the selected first k principal components] for each simulation scene. v 1, v 2..., v k Then, the projection matrix W of each simulation scene is combined to finally obtain the low-dimensional feature dataset matrix ℾ (m×k) of all simulation scenes.

[0048] Assume that the projection matrix remains unchanged throughout the dimensionality reduction and reconstruction process.

[0049] S400: The low-dimensional feature prediction module uses the low-dimensional feature dataset matrix obtained by dimensionality reduction to train the radial basis function neural network model.

[0050] Before training, a radial basis function neural network model is first constructed. The architecture of this model mainly includes an input layer, hidden layers, and an output layer; among which, The input layer is responsible for receiving new variable values ​​of the set design parameters during prediction, thereby forming a new urban design parameter vector, which serves as the input to the radial basis function neural network model. During training, a number of sample data are selected from the standardized data matrix Z using the K-means clustering algorithm, thereby forming an urban design parameter vector, which also serves as the input to the radial basis function neural network model. These sample data are the standardized grid node data and their scene parameters at K different locations in the standardized data matrix Z. The hidden layer contains several nodes and is responsible for mapping the urban design parameter vector input from the input layer to a low-dimensional space through nonlinear transformation, thereby achieving data dimensionality reduction and obtaining the predicted value of the principal component score vector. The output layer is responsible for outputting the predicted values ​​of the principal component score vectors obtained from the hidden layer, forming the principal component score vectors, which serve as the output of the radial basis function neural network model.

[0051] During training, the input layer first uses the K-means clustering algorithm to select new grid node data and their scene parameters at K different locations within the target area from the standardized data matrix Z as new sample data, thereby forming a training urban design parameter vector as input; Then, the radial basis function neural network model is trained using a low-dimensional feature dataset matrix, allowing the hidden layer to learn the nonlinear mapping relationship from the urban design parameter vector to the principal component score vector; Finally, the hidden layer can accurately output the predicted values ​​of the low-dimensional principal component scores based on the urban design parameters, and the output layer outputs k principal component score vectors.

[0052] S500. For new design scenarios where the design parameters remain unchanged and only the variables change within the target area, the low-dimensional feature prediction module uses a trained radial basis function neural network model to make predictions and obtain the predicted values ​​of the low-dimensional new principal component scores of the target area under the new urban design parameters.

[0053] In this step, the specific method for predicting the new design scenario for the target area is as follows: First, set a new variable P1 for each design parameter item within the target area. x P2 x ,…, Pn x This forms a new urban design parameter vector as input; Then, the new urban design parameter vector input by the radial basis function neural network model is used to predict the low-dimensional new principal component score vector of the target area under the new urban design parameters. Finally, the radial basis function neural network model outputs the predicted value of the new principal component score vector of the target area in k dimensions (low dimension) under the new urban design parameters. The specific calculation formula is as follows: (8); (9); (10); In formulas (8)-(10), x p This is the p-th input sample data. c i It is the center point of the i-th grid node. h It is the number of nodes in the hidden layer. n It is the number of output samples. It is the loss function.

[0054] The selected K samples from different locations must be representative of the global data of the target area.

[0055] S600: The high-resolution thermal environment reconstruction module uses the projection matrix W and PCA reconstruction algorithm to reconstruct the predicted values ​​of the k-dimensional new principal component score vector output by the radial basis function neural network model, ultimately restoring the complete high-resolution spatial distribution map of the temperature and wind speed fields of the target area under the new design scenario; the specific calculation formula is as follows: (11); (12); In formulas (11) and (12), Zre represents the reconstructed normalized matrix, and ℾ represents the low-dimensional feature dataset matrix. Xre represents the transpose of the projection matrix, and Xre represents the matrix used to reconstruct the high-dimensional dataset. σ denoted by , where represents the sample standard deviation and μ represents the sample mean.

[0056] S700: The multi-objective optimization decision module first calculates the average temperature and average wind speed of the target area under the new design scenario based on the complete high-resolution temperature and wind speed field spatial distribution map of the target area under the new design scenario obtained by reconstruction. Then, by combining construction and maintenance costs, a comprehensive evaluation index system is constructed; Next, the entropy weight method is used to determine the objective weights of each indicator in the comprehensive evaluation index system; Finally, the scores of the comprehensive evaluation index system for all different design scenarios are calculated, and the optimal urban design parameter scheme is recommended based on the scores.

[0057] The following section uses a community measuring 430m long and 240m wide as an example to explain in detail the urban design parameter optimization method based on high-resolution thermal environment rapid prediction in this application.

[0058] Step 1) First, the buildings and green spaces in the above-mentioned community are taken as optimizable urban design parameters. Then, the building height and green coverage rate are taken as variables. Three types of building heights (7m low-rise buildings, 24m mid-rise buildings and 71m high-rise buildings) and ten types of green coverage rates (0%, 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%) are set. Finally, according to the full factorial design method, a total of 30 simulation scenarios (3×10) are formed. That is, the number of simulation scenarios is equal to the product of the number of variables of each urban design parameter.

[0059] Step 2) First, use ANSYS Fluent software to perform CFD (Computational Fluid Dynamics) simulations on all 30 simulation scenarios. By performing geometric modeling, mesh generation, and thermal environment simulation in sequence, CFD simulation data of all N mesh nodes in each simulation scenario are extracted, namely mesh node data [x, y, z, T, U].

[0060] Each grid node contains spatial coordinates (x, y, z) describing the geometric position of the grid node at a plane 1.5 meters above the ground, as well as air temperature data T and wind speed data U describing the thermal environment characteristics of the grid node at a plane 1.5 meters above the ground.

[0061] Then, all grid node data are organized, and grid node data belonging to the same simulation scenario are combined to form 30 data matrices with 5×N dimensions, namely high-resolution thermal environment data matrices.

[0062] Finally, all 30 high-resolution thermal environment data matrices are combined to form an original high-dimensional dataset matrix with dimensions of 30×5×N, namely the high-resolution thermal environment database X (30×5×N).

[0063] Step 3) First, the high-resolution thermal environment database is standardized using the Z-score method to obtain the standardized data matrix Z. Then, PCA analysis is performed on the standardized data matrix Z to determine that when k=4, the cumulative variance contribution rate reaches 98.5%. Therefore, the first four principal components are selected to construct the projection matrix W= [ v 1, v 2, v 3 , v 4This process compresses each of the original high-dimensional dataset matrices with dimensions 30×5×N into a corresponding 4-dimensional principal component score vector. Finally, the projection matrices W of each simulated scene are combined to obtain the low-dimensional feature dataset matrix ℾ (30×4) of all simulated scenes.

[0064] Step 4) Construct a radial basis function neural network model. First, the input layer of the radial basis function neural network model uses the K-means clustering algorithm to select grid node data and scene parameters at 5 different locations in the community from the standardized data matrix Z as sample data, thus forming a training urban design parameter vector as input (e.g., [x1, y1, z1, T1, U1], [x2, y2, z2, T2, U2], [x3, y3, z3, T3, U3], [x4, y4, z4, T4, U4], [x5, y5, z5, T5, U5] and [P11, P12,…, P1…). q1 , P21, P22,…, P2 q2 ,…,Pn1, Pn2,…,Pn qn Then, the radial basis function neural network model is trained using a low-dimensional feature dataset matrix ℾ, allowing the model's hidden layers to learn a non-linear mapping relationship from the urban design parameter vector to the principal component score vector, until the model can accurately output the predicted low-dimensional principal component scores based on the urban design parameters. See [link to documentation]. Figure 2 As shown, the output layer of the trained radial basis function neural network model consists of four principal component score vectors.

[0065] Step 5) When predicting the thermal environment under a new scenario within the community (design parameters remain unchanged, only their variables are changed) (e.g., a mid-rise building with a height of 20 meters and a green coverage rate of 65%), first set the variable values ​​of the design parameters to form a new urban design parameter vector as input (e.g., [P1]). i =20, P2 j =65%]), and then the radial basis function neural network model predicts the input new urban design parameter vector to obtain the predicted value of the 4-dimensional new principal component score vector of the community under the new urban design parameters and outputs it.

[0066] Note: Since the newly input scene parameters are a scene that has not been CFD simulated, we cannot know the temperature and velocity values ​​under the unknown scene. Therefore, in the prediction, this invention uses the temperature and velocity values ​​of the benchmark scene (the scene with 0% green coverage) instead.

[0067] Step 6) Using the PCA projection matrix W and the PCA reconstruction algorithm, the predicted values ​​of the new 4-dimensional principal component score vectors output by the radial basis function neural network model are reconstructed by increasing their dimensionality. See [link to relevant documentation]. Figure 3 As shown, the complete high-resolution spatial distribution map of temperature and wind speed fields of the target area under the new design scenario (building height 20 meters, green coverage rate 65%) was finally restored.

[0068] Step 7) Based on the complete high-resolution temperature and wind speed spatial distribution map of the target area under the reconstructed new design scenario (building height 20 meters, green coverage 65%), calculate the average temperature and average wind speed of the target area under the new design scenario (building height 20 meters, green coverage 65%). Then, combined with construction and maintenance costs, construct a comprehensive evaluation index system. Then, using the entropy weight method, determine the weights of each temperature, wind speed, and construction and maintenance cost in the comprehensive evaluation index system to be 0.39, 0.8, and 0.33, respectively.

[0069] Step 8) Finally, calculate the score of the comprehensive evaluation index system for all green coverage rates (10%~90%) within a 20-meter building height. See [link / reference]. Figure 4 As shown, the optimal green coverage rate for this 20-meter-high mid-rise building community was determined to be 32% based on the scores.

[0070] This invention creatively proposes a method to first reduce the dimensionality of high-resolution CFD data to low-dimensional data (using PCA to reduce the amount of training data, thereby reducing time costs), then train the model for prediction, and finally perform high-resolution upscaling on the predicted data (using PCA to ensure the generation of high-resolution thermal environments). Therefore, the main innovation of this invention lies in the idea and method of the dimensionality reduction + upscaling process, which can both reduce the prediction cost of machine learning and ensure the high-resolution requirements of thermal environments.

[0071] The present invention also provides a computer device, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus, and the memory is used to store at least one executable instruction, wherein the executable instruction causes the processor to perform the operation corresponding to the above-described method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment.

[0072] The present invention also provides a computer-readable storage medium storing at least one executable instruction that causes a processor to perform the operations corresponding to the above-described method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment, characterized in that, Includes the following steps: S100. Different urban design parameters and their variables are proposed for the infrastructure form in the target area, and several simulation scenarios are designed accordingly. Each simulation scenario has scenario parameters containing its own corresponding urban design parameters and their variable values. S200. Perform CFD simulations on all the simulation scenarios in sequence to obtain the CFD simulation data of each grid node in all the simulation scenarios, and use this to construct the original high-dimensional dataset matrix, that is, obtain the high-resolution thermal environment database. S300. By performing principal component analysis on the high-resolution thermal environment database, the original high-dimensional dataset matrix is ​​projected into a low-dimensional space to obtain a dimensionality-reduced low-dimensional feature dataset matrix. S400. The radial basis function neural network model is trained using the low-dimensional feature dataset matrix, so that the model learns the nonlinear mapping relationship from the urban design parameter vector to the principal component score vector, until the model can accurately output the predicted value of the low-dimensional principal component score according to the urban design parameters. S500: The trained radial basis function neural network model is used to predict new design scenarios in the target area where the design parameters remain unchanged and only the variables change, thereby obtaining the predicted value of the low-dimensional new principal component score of the target area under the new urban design parameters. S600. Using the PCA reconstruction algorithm, the predicted values ​​of the low-dimensional new principal component score vector are reconstructed to restore the complete high-resolution temperature field and wind speed field spatial distribution map of the target area under the new design scenario. S700. Based on the high-resolution temperature and wind speed field spatial distribution maps, combined with average temperature, wind speed, and construction and maintenance costs, a comprehensive evaluation index system is constructed, and the optimal urban design parameter scheme is recommended accordingly.

2. The method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment according to claim 1, characterized in that, In step S100, the method for designing the simulated scenario is specifically as follows: First, based on the infrastructure morphology within the target area, several optimizable urban design parameters are proposed, including buildings, roads, waterways, and green spaces. Then, variables are designed based on the proposed urban design parameters, and several variable values ​​are selected for each of the proposed urban design parameters. Finally, m simulation scenarios are formed according to the full factorial design method. The number of simulation scenarios m is equal to the product of the number of variables of each of the urban design parameter items, and m is a positive integer. The variable values ​​of all the city design parameter items in each simulation scenario together constitute the scenario parameters of that simulation scenario.

3. The method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment according to claim 2, characterized in that, The specific method for establishing the variables of the urban design parameter items is as follows: When a building within the target area is selected as one of the urban design parameters, the building height and / or building density are first set as variables, and then several different building height values ​​and / or different building density values ​​are selected as the variable values ​​of the urban design parameter. When a road within the target area is selected as one of the urban design parameters, the road width and / or road area are first set as variables, and then several different road width values ​​and / or different road density values ​​are selected as the variable values ​​of the urban design parameter. When the water area within the target area is selected as an urban design parameter, the water area and / or water density are first set as variables, and then several different water area values ​​and / or different water density values ​​are selected as variable values ​​for that urban design parameter. When greening within the target area is selected as an urban design parameter, the location of the greening and / or the green coverage rate are first set as variables, and then several different greening locations and / or different green coverage rate values ​​are selected as variable values ​​for that urban design parameter.

4. The method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment according to claim 1, characterized in that, In step S200, the method for constructing the high-resolution thermal environment database is specifically as follows: First, using computational fluid dynamics, geometric modeling, mesh generation, and thermal environment simulation are performed sequentially on all m simulation scenarios. By generating multiple high-resolution thermal environment scenarios covering different scenario parameters, CFD simulation data of all N mesh nodes in each simulation scenario are obtained and denoted as mesh node data, where m and N are both positive integers. Then, all the grid node data are organized, and the grid node data belonging to the same simulation scenario are combined to form m data matrices with a dimension of 5×N, which are denoted as high-resolution thermal environment data matrices. Each simulation scenario corresponds to one high-resolution thermal environment data matrix. Finally, all m high-resolution thermal environment data matrices are combined to form an original high-dimensional dataset matrix with dimensions m×5×N, which is the high-resolution thermal environment database.

5. The method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment according to claim 4, characterized in that, Each of the grid node data contains 5 characteristic parameters, specifically including spatial coordinate data (x, y, z) used to describe the geometric position of the grid node at a plane 1.5 meters above the ground, and air temperature data T and wind speed data U used to describe the thermal environment characteristics of the grid node at a plane 1.5 meters above the ground. Therefore, the grid node data is denoted as [x, y, z, T, U].

6. The method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment according to claim 1, characterized in that, In step S300, the specific steps of the principal component analysis are as follows: S301. The original high-dimensional dataset matrix is ​​preprocessed using the Z-score method to obtain a standardized data matrix; S302. Based on the obtained standardized data matrix, perform covariance calculation to obtain the covariance matrix of the standardized data matrix; S303. Based on the obtained covariance matrix, perform eigenvalue decomposition to obtain several eigenvalues ​​and eigenvectors. Then, arrange the eigenvalues ​​in descending order and define the corresponding eigenvectors as principal components. S304. First, calculate the variance contribution rate of each principal component. Then, using the obtained variance contribution rate of each principal component, select the first k principal components in sequence to calculate the cumulative explained variance contribution rate, where k is a positive integer. Finally, when the calculated cumulative explained variance contribution rate is ≥95%, stop the calculation and select the k principal components corresponding to the current cumulative explained variance contribution rate. S305. Construct a projection matrix belonging to each of the simulated scenarios using the selected first k principal components, and then combine the projection matrices of each of the simulated scenarios to finally obtain the low-dimensional feature dataset matrix of all the simulated scenarios.

7. The method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment according to claim 6, characterized in that, In step S400, the architecture of the radial basis function neural network model includes: The input layer is responsible for receiving new variable values ​​of the set design parameters during prediction, thereby forming a new urban design parameter vector, which serves as the input to the radial basis function neural network model. During training, a number of sample data are selected from the standardized data matrix using the K-means clustering algorithm, thereby forming a training urban design parameter vector, which also serves as the input to the radial basis function neural network model. The sample data are the standardized grid node data and their scene parameters at K different locations in the standardized data matrix. The hidden layer, containing several nodes, is responsible for mapping the urban design parameter vector input from the input layer to a low-dimensional space through nonlinear transformation, thereby achieving data dimensionality reduction and obtaining the predicted value of the principal component score vector. The output layer is responsible for outputting the predicted values ​​of the principal component score vectors obtained from the hidden layer, forming the principal component score vectors as the output of the radial basis function neural network model.

8. The method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment according to claim 7, characterized in that, In step S400, the training method for the radial basis function neural network model is as follows: First, the input layer uses the K-means clustering algorithm to select new grid node data and their scene parameters at K different locations within the target area from the standardized data matrix as new sample data, thereby forming a training urban design parameter vector as input; Then, the radial basis function neural network model is trained using the low-dimensional feature dataset matrix, so that the hidden layer learns the nonlinear mapping relationship from the urban design parameter vector to the principal component score vector; Finally, the radial basis function neural network model can accurately output the predicted values ​​of low-dimensional principal component scores based on the urban design parameters, and the output layer is used to generate k principal component score vectors.

9. The method for optimizing urban design parameters based on rapid prediction of high-resolution thermal environment according to claim 1, characterized in that, In step S700, the method for recommending the optimal urban design parameter scheme is specifically as follows: First, based on the complete high-resolution temperature and wind speed field spatial distribution map of the target area under the new design scenario obtained by reconstruction, calculate the average temperature and average wind speed of the target area under the new design scenario; Then, the average temperature and average wind speed of the target area under the new design scenario are combined with the construction and maintenance costs to construct a comprehensive evaluation index system; Next, the entropy weight method is used to determine the objective weight of each indicator in the comprehensive evaluation index system; Finally, the scores of the comprehensive evaluation index system for all different design scenarios are calculated, and the optimal urban design parameter scheme is recommended based on the scores.

10. A system for implementing the urban design parameter optimization method based on rapid prediction of high-resolution thermal environment as described in any one of claims 1-9, characterized in that, It includes a high-resolution thermal environment database module, a high-dimensional data dimensionality reduction and feature extraction module, a low-dimensional feature prediction module, a high-resolution thermal environment reconstruction module, and a multi-objective optimization decision-making module; among which, The high-resolution thermal environment database module is responsible for proposing different urban design parameters and variables for the infrastructure form in the target area, and designing several simulation scenarios based on these parameters. Then, CFD simulations are performed on all the simulation scenarios in sequence to obtain the CFD simulation data of each grid node in all the simulation scenarios. Based on this, the original high-dimensional dataset matrix is ​​constructed, which is the high-resolution thermal environment database. The high-dimensional data dimensionality reduction and feature extraction module is responsible for performing principal component analysis on the high-resolution thermal environment database, projecting the original high-dimensional dataset matrix into a low-dimensional space, and obtaining a dimensionality-reduced low-dimensional feature dataset matrix. The low-dimensional feature prediction module is responsible for training the radial basis function neural network model using the low-dimensional feature dataset matrix, enabling the model to learn the nonlinear mapping relationship from the urban design parameter vector to the principal component score vector, until the model can accurately output the predicted value of the low-dimensional principal component score based on the urban design parameters. On the other hand, it is responsible for using the trained radial basis function neural network model to predict the new design scenario of the target area, and obtaining the predicted value of the low-dimensional new principal component score of the target area under the new urban design parameters. The high-resolution thermal environment reconstruction module is responsible for using the PCA reconstruction algorithm to reconstruct the predicted values ​​of the low-dimensional new principal component score vectors in a higher dimension, thereby restoring the complete high-resolution temperature field and wind speed field spatial distribution map of the target area under the new design scenario. The multi-objective optimization decision-making module is responsible for constructing a comprehensive evaluation index system based on the high-resolution temperature and wind speed field spatial distribution maps, combined with average temperature, wind speed, and construction and maintenance costs, and recommending the optimal urban design parameter scheme accordingly.

Images