Urban-based green building microclimate regulation emission reduction and sink increase method

Abstract

The application relates to the field of building physical environment simulation and energy-saving control technology, and discloses a green building microclimate regulation emission reduction and carbon sink increase method based on a city, which comprises the following steps: obtaining point cloud data and environmental parameters, solving an initial heat flow path, and calculating a directional velocity component and an air flow influence weight; using a pre-trained deep convolutional neural network to interactively deduce the directional velocity component and the initial heat flow path to obtain a predicted heat distribution, and comparing to obtain a deviation value matrix; performing layout optimization to obtain adjusted layout parameters, and performing boundary calculation to obtain an optimized heat flow path and a velocity vector field; constructing a non-equilibrium region and performing boundary correction processing to obtain a heat transfer difference; inputting a preset heat and humidity coupling equation to solve a maximum fluctuation amplitude; and using a pre-trained graph neural network to topologically deduce the optimized heat flow path to obtain an interaction model. The method realizes high-precision microclimate simulation.

Description

Technical Field

[0001] This invention relates to the field of building physical environment simulation and energy-saving control technology, and in particular to a method for green building microclimate regulation, emission reduction and carbon sequestration enhancement based on cities. Background Technology

[0002] Currently, against the backdrop of increasingly intertwined urban development and environmental protection, green building, as an important path to achieving sustainable development, demonstrates irreplaceable value. Especially under the goals of urban microclimate regulation and emission reduction and carbon sequestration, how to optimize the interaction between buildings and their surrounding environment through technological means not only concerns the quality of life of urban residents but also directly affects energy consumption and ecological balance.

[0003] In existing technologies, building microclimate assessments typically rely on computer-aided design software or conventional static environmental measurement tools. These methods often focus solely on passive energy-saving design of the building itself, such as replacing exterior wall insulation materials, or improving single environmental factors, such as local wind speed or static sunlight. However, as part of the urban environment, the shape, materials, and layout of a building directly alter the surrounding airflow distribution and heat transfer, which in turn dynamically affects the building's energy consumption. The aforementioned existing technologies neglect the complex dynamic balance between buildings and the urban microclimate. Due to the lack of in-depth flow-heat coupling analysis, it is difficult to accurately capture and simulate heat flow paths in actual physical environments, let alone reveal the mutual constraints between airflow guidance and heat distribution. Specifically, the uncertainty of heat flow paths makes the effect of airflow guidance difficult to predict, while deviations in airflow guidance exacerbate the uneven heat distribution, creating a double technical dilemma.

[0004] Existing technologies have limitations in achieving high-precision, dynamically adaptive microclimate regulation and emission reduction / sinking in green buildings. Summary of the Invention

[0005] This invention provides a method for green building microclimate regulation, emission reduction and carbon sequestration enhancement based on urban conditions, in order to solve the problem in the prior art that it is difficult to achieve high-precision, dynamic, adaptive microclimate regulation and emission reduction and carbon sequestration enhancement for green buildings.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a method for green building microclimate regulation, emission reduction, and carbon sequestration enhancement based on urban areas, comprising: The point cloud data of the building and the surrounding environmental parameters are acquired, and the initial heat flow path is obtained by solving the point cloud data and the environmental parameters using computational fluid dynamics algorithms. Based on the initial heat flow path, convective heat transfer calculations are performed to obtain the directional velocity component and the airflow influence weight; If the airflow influence weight is greater than the preset weight threshold, a pre-trained deep convolutional neural network is used to interactively deduce the directional velocity component and the initial heat flow path to obtain the predicted heat distribution, and the predicted heat distribution is compared with the initial heat flow path to obtain the deviation value matrix. The layout is optimized based on the deviation matrix and the preset component constraints to obtain the adjusted layout parameters, and the boundary calculation is performed using the adjusted layout parameters to obtain the optimized heat flow path and velocity vector field. The transient energy fluctuation rate is determined based on the velocity vector field, and when the transient energy fluctuation rate is greater than a preset fluctuation threshold, a non-equilibrium region is constructed using the velocity vector field. The boundary correction process of the non-equilibrium region is performed using the optimized heat flow path to obtain the heat transfer difference. The ambient temperature and humidity are obtained, and the heat transfer difference and the ambient temperature and humidity are input into a preset heat and humidity coupling equation for solution to obtain the maximum fluctuation amplitude. If the maximum fluctuation amplitude is less than the preset adjustment threshold, the optimized heat flow path is topologically deduced using a pre-trained graph neural network to obtain an interaction model.

[0007] Secondly, this invention provides a green building microclimate regulation, emission reduction, and carbon sequestration enhancement system based on urban areas, comprising: The data processing module is used to acquire point cloud data of the building and surrounding environmental parameters, and use computational fluid dynamics algorithms to solve the point cloud data and environmental parameters to obtain the initial heat flow path; The convection calculation module is used to perform convection heat transfer calculation based on the initial heat flow path to obtain the directional velocity component and the airflow influence weight; The feature inference module is used to perform interactive inference on the directional velocity component and the initial heat flow path to obtain a predicted heat distribution if the influence weight of the airflow is greater than a preset weight threshold, and compare the predicted heat distribution with the initial heat flow path to obtain a deviation value matrix. The boundary calculation module is used to perform layout optimization based on the deviation value matrix and preset component constraints to obtain adjusted layout parameters, and to use the adjusted layout parameters to perform boundary calculation to obtain optimized heat flow path and velocity vector field. The energy fluctuation module is used to determine the transient energy fluctuation rate based on the velocity vector field, and when the transient energy fluctuation rate is greater than a preset fluctuation threshold, to construct a non-equilibrium region using the velocity vector field, and to perform boundary correction processing on the non-equilibrium region using the optimized heat flow path to obtain the heat transfer difference; The interactive modeling module is used to acquire the ambient temperature and humidity, and input the heat transfer difference and the ambient temperature and humidity into a preset heat and humidity coupling equation for solving to obtain the maximum fluctuation amplitude. If the maximum fluctuation amplitude is less than the preset adjustment threshold, the pre-trained graph neural network is used to perform topological deduction on the optimized heat flow path to obtain the interaction model.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention uses a pre-trained deep convolutional neural network to interactively deduce the directional velocity component and the initial heat flow path to obtain the predicted heat distribution, and then compares it with the initial heat flow path to obtain the deviation value matrix. This breaks through the bottleneck of traditional linear analysis models in extracting the characteristics of strong interference airflow when dealing with complex urban microclimates. It can accurately capture the micro-modulation effect of airflow disturbance on the heat transfer path, such as the deep physical interaction mode of vortex heat retardation effect, and realize the high-fidelity identification of nonlinear spatial deviation under complex flow-heat coupling effect, thereby improving the accuracy of building heat distribution prediction under microclimate environment.

[0009] (2) This invention uses the deviation value matrix and component constraints to generate adjusted layout parameters to reset the preset calculation boundary conditions, and further calculates the transient energy fluctuation rate of the velocity vector field. For abnormal nodes exceeding the threshold, non-equilibrium regions are constructed to perform local boundary correction and recalculation. The passive static environment calculation is transformed into active physical boundary closed-loop optimization. This not only effectively corrects the macroscopic guidance direction of the global flow field, but also accurately suppresses the transient energy oscillations caused by turbulence in the local area. It realizes the dynamic adaptive feedback of building space form to microclimate changes, completely eliminates the risk of local heat accumulation caused by ignoring transient fluctuations, and improves the heat dissipation efficiency and airflow organization quality of the building fluid-solid interface.

[0010] (3) This invention verifies compliance by inputting the recalculated heat transfer difference and the ambient temperature and humidity into the heat-humidity coupling equation to solve the maximum fluctuation amplitude. After meeting the standard, it uses a pre-trained graph neural network to perform topological deduction on the optimized heat flow path to obtain an interaction model. This upgrades the traditional isolated load assessment to a full-cycle dynamic physical process map based on network topology evolution. It profoundly reveals the spatiotemporal constraints of how heat flow interferes with airflow guidance and how airflow reverses and reshapes heat distribution. It realizes the leap from single-point static monitoring to global dynamic evolution deduction for building microclimate regulation. It provides a highly reliable scientific decision-making model for refined air volume allocation and emission reduction and sinking control, and reduces the comprehensive operating energy consumption of high-density building clusters in cities. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the process of the green building microclimate regulation, emission reduction and carbon sequestration enhancement method based on the first embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a green building microclimate regulation, emission reduction and sequestration system based on cities, provided in the second embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Reference Figure 1 The first embodiment of the present invention provides a method for green building microclimate regulation, emission reduction and carbon sequestration enhancement in cities, including the following steps: S11, acquire the point cloud data of the building and the surrounding environmental parameters, and use computational fluid dynamics algorithm to solve the point cloud data and the environmental parameters to obtain the initial heat flow path; S12, perform convective heat transfer calculation based on the initial heat flow path to obtain the directional velocity component and airflow influence weight; S13, if the airflow influence weight is greater than the preset weight threshold, the predicted heat distribution is obtained by interactively deducing the directional velocity component and the initial heat flow path using a pre-trained deep convolutional neural network, and the predicted heat distribution is compared with the initial heat flow path to obtain the deviation value matrix. S14. Based on the deviation value matrix and the preset component constraints, the layout optimization is performed to obtain the adjusted layout parameters, and the boundary calculation is performed using the adjusted layout parameters to obtain the optimized heat flow path and velocity vector field. S15, determine the transient energy fluctuation rate based on the velocity vector field, and when the transient energy fluctuation rate is greater than a preset fluctuation threshold, construct a non-equilibrium region using the velocity vector field, and perform boundary correction processing on the non-equilibrium region using the optimized heat flow path to obtain the heat transfer difference; S16, acquire the ambient temperature and humidity, and input the heat transfer difference and the ambient temperature and humidity into the preset heat and humidity coupling equation for solution to obtain the maximum fluctuation amplitude. If the maximum fluctuation amplitude is less than the preset adjustment threshold, use the pre-trained graph neural network to perform topological deduction on the optimized heat flow path to obtain the interaction model.

[0014] In step S11, point cloud data of the building and surrounding environmental parameters are acquired, and an initial heat flow path is obtained by solving the point cloud data and environmental parameters using a computational fluid dynamics algorithm, including: Acquire point cloud data and extract surface features from the point cloud data to obtain material layout properties; Construct a building entity model based on the aforementioned material layout properties; The building entity model is discretized based on environmental parameters to generate a numerical computation grid space; The velocity vector field and temperature scalar field are obtained by solving the numerical computation grid space using computational fluid dynamics algorithms. The instantaneous heat flux distribution is calculated based on the velocity vector field and the temperature scalar field. The initial heat flux path is obtained by extracting the spatial evolution trajectory of the instantaneous heat flux distribution.

[0015] In one implementation, this embodiment uses airborne or ground-based lidar equipment to perform a three-dimensional scan of the target area and acquire point cloud data.

[0016] It should be noted that the point cloud data includes the three-dimensional coordinates, reflection intensity values, and three-channel color values ​​of spatially discrete points. This embodiment extracts surface features from the point cloud data to obtain material layout attributes. Specifically, the reflection intensity value and three-channel color values ​​of each discrete point in the point cloud data are extracted to construct a multi-dimensional feature vector. This multi-dimensional feature vector is then input into a pre-trained support vector machine classification model, which outputs a material category label corresponding to each discrete point. The material category label explicitly covers three categories: glass, concrete, and vegetation.

[0017] It is worth noting that the construction process of the pre-trained support vector machine classification model is as follows: a large amount of point cloud feature data of known material surfaces is collected as training samples, and a radial basis function kernel is adopted. The penalty parameter and the radial basis kernel function width parameter are determined by searching for the optimal combination of parameters on the validation set through a grid search algorithm combined with a ten-fold cross-validation mechanism. The model is then optimized by solving the Lagrange dual function that maximizes the classification margin.

[0018] In one implementation, this embodiment constructs a building entity model based on the material layout attributes. This embodiment segments the point cloud data into regions based on the material category labels, performs a Poisson surface reconstruction algorithm on the segmented point cloud sets to generate three-dimensional geometric envelopes, and assigns corresponding physical property parameters to different geometric envelopes. Preset porosity and transpiration heat dissipation coefficients are assigned to the geometric envelopes labeled "vegetation," preset thermal conductivity is assigned to the geometric envelopes labeled "glass," and preset surface roughness values ​​are assigned to the geometric envelopes labeled "concrete." The building entity model is generated by combining the three-dimensional geometric envelopes and their corresponding physical property parameters. It should be noted that the preset porosity, transpiration heat dissipation coefficient, thermal conductivity, and surface roughness values ​​are all determined by retrieving corresponding physical measurement values ​​from a standard building material and botanical property parameter database.

[0019] In one implementation, this embodiment discretizes the building entity model using environmental parameters to generate a numerical computation mesh space. Environmental parameters, including dry-bulb temperature, prevailing wind direction vector, and wind speed, are obtained. Based on these parameters, a virtual fluid domain completely enclosing the building entity model is constructed. This embodiment employs a polyhedral meshing algorithm to spatially partition the virtual fluid domain.

[0020] It should be noted that, for the local space containing the vegetation envelope with the aforementioned porosity, this embodiment performs a local mesh subdivision operation based on a preset mesh refinement scale to generate the numerical calculation mesh space. The preset mesh refinement scale is obtained by performing an independent mesh test on a standard porous medium and selecting the maximum mesh feature length when the iteration difference rate of the fluid pressure loss calculation results is less than 5%.

[0021] In one implementation, this embodiment utilizes computational fluid dynamics algorithms to solve the numerical computational grid space to obtain the velocity vector field and temperature scalar field. This embodiment sets the environmental parameters as the inlet boundary conditions of the numerical computational grid space, employs the steady-state incompressible Reynolds-averaged Navier-Stokes equations as the governing equations, and combines them with a shear stress transport turbulence model. Iterative numerical integration calculations are then performed using a finite volume solver. When both the continuity residual and momentum residual of the governing equations decrease below a preset convergence threshold, the system stops iterative calculations and outputs the three-dimensional velocity vector and absolute temperature value of each grid node in the numerical computational grid space, which are then combined to form the velocity vector field and the temperature scalar field, respectively. The preset convergence threshold is set to 0.0001, a value derived based on the discrete error control criteria in numerical analysis.

[0022] In one implementation, this embodiment calculates the instantaneous heat flux distribution based on the velocity vector field and the temperature scalar field. This embodiment extracts the velocity modulus of each grid node in the velocity vector field, and calculates the convective heat transfer coefficient using the empirical formula for the nonlinear exponential product of the Reynolds number and Prandtl number calculated based on the characteristic length of the grid in fluid dynamics, i.e., the Nusselt number criterion correlation. Simultaneously, it extracts the temperature gradient values ​​between adjacent grid nodes based on the temperature scalar field. The convective heat transfer coefficient is multiplied by the temperature gradient values ​​to obtain the convective heat flux density, which is then added to the radiative heat flux density calculated according to the Stefan-Boltzmann law to obtain the total heat flux vector of each grid node. This vector is then collected by traversing all grid nodes to generate the instantaneous heat flux distribution.

[0023] In one implementation, this embodiment extracts the spatial evolution trajectory of the instantaneous heat flux distribution to obtain the initial heat flow path. This embodiment sets grid nodes with a total heat flux vector magnitude greater than a preset heat source threshold as the starting points for tracking, and uses a fourth-order Runge-Kutta numerical integration algorithm to perform spatial step integration calculations along the vector direction of the instantaneous heat flux distribution.

[0024] It should be noted that the spatial step size for the spatial step integral calculation is determined dynamically based on the Courant-Friedrich-Lévy stability condition (CFL condition) in numerical analysis, combined with the local maximum velocity value in the velocity vector field and the characteristic length of the grid cells, to prevent distortion due to integration exceeding limits. The discrete coordinate points generated by each integral step size are sequentially connected to form a continuous curve coordinate set in three-dimensional space, outputting the initial heat flow path. The preset heat source threshold is obtained by statistically analyzing the measured heat flux values ​​of various building exterior surfaces, including roofs and east, south, and west-facing walls, during the direct sunlight period on typical summer weather days in the target city; the 75th percentile of this set is calculated and calibrated as the preset heat source threshold.

[0025] For example, the acquired environmental parameters are a dry-bulb temperature of 32.5 degrees Celsius, a southeast wind direction vector, and a wind speed of 4.0 meters per second. After computational fluid dynamics solving of the numerical computation grid space, the output velocity vector field shows that the airflow velocity modulus reaches 5.5 meters per second when passing through the building corner, while the velocity modulus drops to 0.8 meters per second in the area passing through the vertical green layer. Simultaneously, the output temperature scalar field shows that the absolute temperature behind the vertical green coverage area is 2.1 degrees Celsius lower than that of the exposed wall surface. The system performs a fourth-order Runge-Kutta numerical integration algorithm on the generated instantaneous heat flux distribution to extract a three-dimensional continuous curve rising from the heated asphalt pavement and ascending along the lower floors of the building. The coordinate set of this curve is the final generated initial heat flux path.

[0026] In step S12, convective heat transfer calculation is performed based on the initial heat flow path to obtain the directional velocity component and airflow influence weight, including: Node parameters are extracted based on the initial heat flow path, and scalar and vector feature separation is performed on the node parameters to obtain the airflow velocity field and airflow direction field; The directional velocity component is obtained by spatial vector fusion of the airflow velocity field and the airflow direction field. By combining the preset air density and the preset specific heat capacity, the convective heat transfer capacity of the grid nodes is calculated using the directional velocity component to obtain the convective heat transfer intensity matrix. Spatial correlation analysis was performed on the convective heat transfer intensity matrix to obtain the airflow guidance coupling coefficient; The airflow influence weights are obtained by solving the airflow guidance coupling coefficient using a multiple linear regression algorithm.

[0027] In one implementation, this embodiment extracts node parameters based on the initial heat flow path and performs scalar and vector feature separation on the node parameters to obtain the airflow velocity field and airflow direction field. This embodiment analyzes the evolution trajectory of the initial heat flow path in three-dimensional space, extracts the fluid dynamics values ​​on the spatial grid nodes as the node parameters, performs feature decomposition on the node parameters, separates the numerical distribution representing the speed of airflow to construct the airflow velocity field, and simultaneously extracts three-dimensional unit vectors indicating the airflow trajectory to construct the airflow direction field.

[0028] It should be noted that in this embodiment, the airflow velocity field and the airflow direction field are fused using spatial vectors to obtain the directional velocity component. A single velocity scalar value is multiplied by a unit direction vector in a three-dimensional Cartesian coordinate system, transforming it into three-dimensional coordinate system data with positive lateral, positive longitudinal, and positive vertical components, and the directional velocity component is then output.

[0029] In one implementation, this embodiment combines a preset air density and a preset specific heat capacity, substitutes the magnitude of the directional velocity component into the empirical correlation of forced convection heat transfer based on Reynolds number and Prandtl number, and performs nonlinear polynomial calculations to obtain the convective heat transfer intensity characterization benchmark for each grid node in space. The convective heat transfer intensity characterization benchmarks of all grid nodes are then arranged and assembled according to their grid coordinate system positions to generate the convective heat transfer intensity matrix. It is worth noting that the preset air density and preset specific heat capacity are determined by retrieving objective lookup values ​​from a standard meteorological property parameter database under local ambient temperature and standard atmospheric pressure conditions.

[0030] It should be noted that in this embodiment, spatial correlation analysis is performed on the convective heat transfer intensity matrix to obtain the airflow guiding coupling coefficient. This embodiment extracts the heat transfer intensity numerical sequences of the target grid node and its neighboring grid nodes at multiple historical steady-state time steps, calculates the Pearson correlation coefficient between these two time sequences, and extracts the absolute value of the Pearson correlation coefficient as the airflow guiding coupling coefficient to quantify the heat transfer correlation strength between the building structure and the surrounding airflow.

[0031] In one implementation, this embodiment uses a multiple linear regression algorithm to calculate the airflow guidance coupling coefficient to obtain the airflow influence weight. This embodiment constructs a multiple linear regression model, inputs the airflow guidance coupling coefficient as an independent variable into the model to perform matrix algebra operations, and outputs the airflow influence weight. This value physically represents the percentage of heat transfer changes caused by airflow guided by the building facade structure relative to the dynamic changes in total local heat transfer.

[0032] Specifically, the airflow guidance coupling coefficient corresponding to each grid node is used as the dependent variable sample, and the three-dimensional spatial coordinates of the node, the local Reynolds number and turbulence intensity at the node are used as independent variables. A multiple linear regression equation is constructed, and the regression coefficients of each independent variable are obtained by fitting with the least squares method. Then, all regression coefficients are weighted and summed, and after minimum-maximum normalization, the airflow influence weight of the node is obtained.

[0033] It is worth noting that the aforementioned multiple linear regression model was constructed by collecting a large amount of historical similar building microclimate environment simulation flow-thermal coupling datasets and fitting the optimal regression coefficients based on the least squares criterion.

[0034] For example, the system identifies an airflow velocity of 3.14 meters per second at a node, oriented 15 degrees east of north with a positive elevation angle. Through spatial vector fusion, the resulting directional velocity components are 0.8 meters per second laterally, 3.0 meters per second longitudinally, and 0.47 meters per second vertically. The system calculates the sum of the squares of the lateral, longitudinal, and vertical components, and then takes the square root of this summation, yielding an airflow velocity of approximately 3.14 meters per second. Combining this with retrieved air property parameters, the system calculates the airflow heat capacity carrying capacity per unit area of ​​the node to be 42 watts per square meter (Kelvin), using this as the benchmark for convective heat transfer intensity at that node, thus forming a local convective heat transfer intensity matrix. Comparing the heat transfer intensity gradient changes between adjacent nodes in the matrix, the airflow guidance coupling coefficient is calculated to be 0.78. Substituting this into a multiple linear regression model, the airflow influence weight is calculated to be 0.65.

[0035] In step S13, if the influence weight of the airflow is greater than a preset weight threshold, a pre-trained deep convolutional neural network is used to interactively deduce the directional velocity component and the initial heat flow path to obtain a predicted heat distribution, and the predicted heat distribution is compared with the initial heat flow path to obtain a deviation value matrix.

[0036] Specifically, a pre-trained deep convolutional neural network is used to interactively extrapolate the directional velocity component and the initial heat flow path to obtain a predicted heat distribution. The predicted heat distribution is then compared with the initial heat flow path to obtain a deviation matrix, including: The directional velocity component and the initial heat flow path are mapped to a preset three-dimensional mesh space; Within the three-dimensional mesh space, the directional velocity component and the initial heat flow path are tensor-concatenated to generate a spatial feature vector; The spatial feature vector is input into a pre-trained deep convolutional neural network, and deep interaction feature matrix is ​​extracted using multi-layer convolution operations. The predicted heat distribution is obtained by decoding and deducing the deep interaction feature matrix using a pre-trained decoder. The predicted heat distribution is compared with the initial heat flow path point by point to obtain the deviation matrix by calculating the difference.

[0037] In one implementation, this embodiment determines whether the airflow influence weight is greater than a preset weight threshold; if the airflow influence weight is greater than the preset weight threshold, the directional velocity component and the initial heat flow path are mapped to a preset three-dimensional mesh space. This embodiment divides the spatial environment into the preset three-dimensional mesh space with a fixed discrete resolution. Based on the one-to-one correspondence of three-dimensional physical coordinates, the three-dimensional numerical features of the directional velocity component and the geometric trajectory point features of the initial heat flow path are assigned and filled into the preset three-dimensional mesh space.

[0038] It is worth noting that the preset weight threshold is obtained by collecting airflow influence weights and corresponding thermal deviation data from historical projects, plotting scatter plots and performing curve fitting, calculating the first derivative of the fitted curve at each data point, sorting all derivative values ​​by size, and taking the average of the airflow influence weights corresponding to the top 5% of derivative values ​​as the preset weight threshold. This method can reliably capture the critical point where airflow influence begins to change significantly.

[0039] It should be noted that in this embodiment, the directional velocity component and the initial heat flow path are aligned and tensor merged and spliced ​​along the channel feature dimension, i.e. the depth dimension, within the three-dimensional mesh space to generate the spatial feature vector represented by a multi-channel data structure.

[0040] In one implementation, this embodiment inputs the spatial feature vector into a pre-trained deep convolutional neural network and extracts a deep interaction feature matrix using multi-layer convolution operations. In this embodiment, the spatial feature vector is fed into the deep convolutional neural network architecture, and by performing sliding multiply-accumulate operations on multi-layer three-dimensional convolutional kernels and mapping with nonlinear activation functions, a multi-dimensional tensor of vortex hysteresis effect and shear flow heat dissipation characteristics is extracted, i.e., the deep interaction feature matrix.

[0041] It is worth noting that the pre-trained deep convolutional neural network adopts a three-dimensional U-Net network topology architecture, including a three-layer downsampling encoder structure and a three-layer upsampling decoder structure. Its core convolutional operation uses a three-dimensional sliding convolution kernel with a size of 3x3x3. During the model training phase, the initial learning rate is set to 0.001, and the batch size is set to 16. The model collects heterogeneous coupled data generated by cross-validation of large-scale wind tunnel physics experiments and high-precision computational fluid dynamics simulations as training samples. Based on the criterion of minimizing the root mean square error between the predicted tensor and the actual physical field distribution tensor, the backpropagation algorithm is used to iteratively update the network node weights.

[0042] It should be noted that this embodiment utilizes a pre-trained decoder to decode and deduce the deep interaction feature matrix to obtain the predicted heat distribution. In this embodiment, the deep interaction feature matrix is ​​input into the deconvolutional layer structure of the decoder, performing feature upsampling and dimensionality expansion operations and cross-layer feature fusion. This maps and restores the high-dimensional hidden layer abstract features to a full-resolution three-dimensional grid space, outputting a spatial temperature field distribution map considering nonlinear hydrodynamic coupling effects, i.e., the predicted heat distribution.

[0043] In one implementation, this embodiment compares the predicted heat distribution with the initial heat flow path point by point to calculate the difference and obtain a deviation matrix. This embodiment traverses the preset three-dimensional grid spatial coordinates, extracts the predicted temperature values ​​output by the predicted heat distribution at each grid node, subtracts these values ​​from the steady-state temperature values ​​of the initial heat flow path at the corresponding nodes, calculates the absolute temperature difference of each node, and then arranges and combines the absolute temperature differences of all grid nodes in a matrix according to the original spatial topology to generate the deviation matrix.

[0044] It should be noted that before the comparison, the temperature values ​​on the initial heat flow path are first diffused to all grid nodes in the preset three-dimensional grid space through the inverse distance weighted interpolation algorithm to generate an initial heat flow path temperature field with the same resolution as the predicted heat distribution, and then a point-by-point temperature comparison is performed.

[0045] For example, when the influence weight of the airflow guiding the airflow (0.65) is determined to be greater than the preset weight threshold of 0.60, the system aligns and splices the trajectory features of the heat flow path in the high-altitude area of ​​the atrium with the directional velocity component of the airflow along the channel dimension within a preset three-dimensional grid space, generating a spatial feature vector. This vector is then input into a pre-trained deep convolutional neural network using a three-dimensional U-Net network topology architecture. After multi-layer sliding convolution operations, a deep interactive feature matrix characterizing the nonlinear coupling features of the inner side of the west glass curtain wall is extracted. Further, an upsampling decoder is used to deduce the predicted heat distribution. The system extracts the steady-state temperature value of the initial heat flow path in the conference room area and the predicted value of the predicted heat distribution at that location. By subtracting these values, a 3.5-degree Celsius temperature rise difference is found at the corner. The system records all calculated differences, including this 3.5-degree Celsius difference, at the corresponding grid coordinate points, generating the final deviation value matrix.

[0046] In step S14, layout optimization is performed based on the deviation value matrix and preset component constraints to obtain adjusted layout parameters, and boundary calculations are performed using the adjusted layout parameters to obtain optimized heat flow paths and velocity vector fields, including: Based on the deviation value matrix, a correction factor is assigned to the preset three-dimensional grid space to generate a spatial correction factor matrix; The spatial correction factor matrix is ​​used to generate shape adjustment instructions within the preset component constraints, and the adjusted layout parameters are generated according to the shape adjustment instructions. The preset computational boundary conditions are reset using the adjusted layout parameters to construct an updated thermodynamic conduction model; The updated thermodynamic conduction model is input into the fluid dynamics solver for global solution to obtain the optimized heat flow path and velocity vector field.

[0047] In one implementation, this embodiment assigns a correction factor to a preset three-dimensional mesh space based on the deviation value matrix, generating a spatial correction factor matrix. This embodiment iterates through each mesh node in the deviation value matrix, extracts the temperature deviation amplitude value of the corresponding node, and constructs a mapping function to convert the temperature deviation amplitude value into the corresponding correction factor value.

[0048] The specific mapping logic is as follows: when the temperature deviation amplitude approaches zero, a base correction factor of 1 is assigned; when the temperature deviation amplitude significantly deviates from zero, a corresponding offset correction factor is assigned according to a preset inverse proportional relationship. The specific calculation logic for this inverse proportional relationship is to subtract the product of the temperature deviation amplitude and a preset scaling factor from the base correction factor of 1 to obtain the offset correction factor value. The correction factor values ​​calculated from all grid nodes are then arranged in a matrix to generate the spatial correction factor matrix. It should be noted that the preset scaling factor is determined by performing multiple linear regression analysis on local airflow offset calibration experimental data of historical buildings of the same type, extracting the slope characteristic value of the fitted straight line.

[0049] It should be noted that in this embodiment, the spatial correction factor matrix is ​​used to generate a form adjustment instruction within a preset component constraint, and the adjusted layout parameters are generated according to the form adjustment instruction. This embodiment identifies abnormal nodes in the spatial correction factor matrix where the correction factor value is not equal to the basic correction factor value, extracts the spatial coordinates corresponding to the abnormal nodes, retrieves physical regions containing the spatial coordinates in a pre-constructed building information model, identifies controlled building components within the physical regions, and reads the preset component constraints corresponding to the controlled building components.

[0050] It is worth noting that the preset component constraints cover the physical limit range and adjustment step size of the adjustable parameters of the component. This constraint data is determined by consulting the building component's factory specifications and physical assembly limitation conditions, and storing them in a relational database.

[0051] This embodiment calculates the target adjustment amount within the preset component constraints based on the correction factor value. The specific logic for calculating the target adjustment amount involves constructing a linear mapping relationship based on the correction factor; that is, the target adjustment amount equals the product of the component's maximum adjustable physical extreme value, the correction factor, and the preset sensitivity coefficient, followed by the addition of the reference installation offset. The preset sensitivity coefficient is an objective physical parameter determined by conducting offline step response tests on adjustable components of the target building, such as shading louvers and actuators, calibrating the linear response slope between the change in mechanical angle and the width of the control command pulse. This embodiment generates the morphological adjustment command containing the component's physical identifier and the target adjustment amount; it updates the original building model structural parameters according to the morphological adjustment command, generating the adjusted layout parameters.

[0052] In one implementation, this embodiment uses the adjusted layout parameters to reset the preset computational boundary conditions and construct an updated thermodynamic conduction model. This embodiment converts the adjusted layout parameters into three-dimensional geometric boundary mesh data, replaces the corresponding geometric entity surfaces in the original fluid dynamics computational domain, and updates the spatial coordinates and angular orientation of the fluid-structure interaction surface, thereby resetting the preset computational boundary conditions. Based on the reset computational boundary conditions and the inherent thermal properties of the building materials, the thermodynamic conduction model, which includes the coupled computational features of convection heat transfer and heat conduction, is reconstructed.

[0053] It should be noted that in this embodiment, the updated thermodynamic conduction model is input into the fluid dynamics solver for global solution to obtain the optimized heat flux path and velocity vector field. This embodiment again invokes the steady-state incompressible control equations and the shear stress transport turbulence model, inputting them into the fluid dynamics solver for numerical iterative calculation. After the residual decreases to a preset convergence threshold, the three-dimensional wind speed components of the global grid nodes are extracted to construct the velocity vector field, and the spatial heat flux vector is extracted and the optimized heat flux path is generated using an integral tracing algorithm.

[0054] For example, in response to the abnormal heat accumulation on the south side of the office area, the system assigns a correction factor of 0.85 to the corresponding grid nodes in that area. The system identifies the shading louvers in this area as adjustable building components and reads their component constraints, such as a maximum adjustable physical limit of 90 degrees. The system performs pure textual logic calculations based on a linear mapping relationship, multiplying 90 degrees, the correction factor of 0.85, and the preset sensitivity coefficient of 0.8 obtained through offline calibration, adding the baseline installation offset of 0 degrees, to calculate a target adjustment of approximately 61 degrees, rounded to 60 degrees by the step size. The system generates a shape adjustment command to adjust the tilt angle of the shading louvers near the exterior window from 45 degrees to 60 degrees, generating adjusted layout parameters. These adjusted layout parameters are used to reset the geometric boundary conditions in the 3D model and reconstruct the thermodynamic conduction model. After inputting the updated model into the fluid dynamics solver for global solution, the airflow velocity in the stagnant region changes from 0.2 meters per second to 0.6 meters per second, generating a new velocity vector field; at the same time, the heat flow trajectory in this region changes to a directional horizontal flow conductor towards the indoor return air vent, generating an optimized heat flow path.

[0055] In step S15, the transient energy fluctuation rate is determined based on the velocity vector field, and when the transient energy fluctuation rate is greater than a preset fluctuation threshold, a non-equilibrium region is constructed using the velocity vector field. The boundary correction process of the non-equilibrium region is performed using the optimized heat flow path to obtain the heat transfer difference.

[0056] The process includes determining the transient energy fluctuation rate based on the velocity vector field, and constructing a non-equilibrium region using the velocity vector field when the transient energy fluctuation rate exceeds a preset fluctuation threshold. Extract the three-dimensional wind speed components of each node in the velocity vector field; The transient energy fluctuation rate is obtained by calculating the kinetic energy variance using the three-dimensional wind speed components. Determine whether the transient energy fluctuation rate is greater than a preset fluctuation threshold; If the transient energy fluctuation rate is greater than the preset fluctuation threshold, the corresponding abnormal node coordinates are extracted from the velocity vector field; The coordinates of the abnormal nodes are used to define the spatial range and construct a non-equilibrium region.

[0057] In one implementation, this embodiment extracts the three-dimensional wind speed components of each node in the velocity vector field; the kinetic energy variance is calculated using the three-dimensional wind speed components to obtain the transient energy fluctuation rate. This embodiment extracts the three-dimensional wind speed component values ​​of a single grid node in the velocity vector field within the current time step and several previous historical time steps to form a time series; the arithmetic mean of each wind speed component value in this time series is calculated; the squared difference between the wind speed component value at each time step in the series and the arithmetic mean is calculated respectively; all squared differences are summed and divided by the total number of time steps to calculate the kinetic energy variance value, which quantifies the dispersion of the fluid micro-particle's motion energy; the kinetic energy variance value is then confirmed as the transient energy fluctuation rate characterizing the local turbulent fluctuation features.

[0058] It should be noted that the velocity vector field is obtained by solving an unsteady computational fluid dynamics algorithm, outputting three-dimensional wind speed component data for multiple consecutive time steps to form a time series. In another implementation, the transient energy fluctuation rate can also be based on the velocity vector field of a single time step, calculating the variance of the wind speed component in the spatial neighborhood of each grid node, characterizing the spatial non-uniformity of local turbulent fluctuation energy.

[0059] It should be noted that this embodiment determines whether the transient energy fluctuation rate is greater than a preset fluctuation threshold. It is worth noting that the preset fluctuation threshold is determined by statistically analyzing historical sensor data from wind tunnel experiments of the target building type under ideal laminar flow conditions, extracting the kinetic energy variance data set of all measuring points under stable airflow organization, and selecting the 95th percentile value using the statistical percentile method. This embodiment performs a logical judgment operation, comparing the currently calculated transient energy fluctuation rate value with the preset fluctuation threshold value.

[0060] In one implementation, if the transient energy fluctuation rate is determined to be greater than the preset fluctuation threshold, the corresponding abnormal node coordinates are extracted from the velocity vector field; the abnormal node coordinates are then used to define a spatial range and construct a non-equilibrium region. In this embodiment, grid nodes that meet the judgment condition are marked as abnormal state points, and the absolute spatial coordinate coefficients of these abnormal state points in the three-dimensional grid space of the velocity vector field are extracted to obtain the abnormal node coordinates; a density-based noisy spatial clustering algorithm, namely the DBSCAN algorithm, is used to perform cluster analysis on all the extracted abnormal node coordinates.

[0061] It should be noted that the preset neighborhood radius is determined by performing nearest neighbor distance statistical graph analysis (K-distance graph) on historical grid point cloud data, selecting the distance value corresponding to the point with the maximum rate of change of the slope of the distance abrupt change curve, i.e., the extreme point of the second derivative; the criterion for determining whether the node density meets the standard is that the number of nodes in the neighborhood is greater than the preset minimum sample number, which is set based on twice the spatial dimension, i.e., the value four. Coordinate clusters whose spatial straight-line distance is less than the preset neighborhood radius and meet the node density standard are spatially merged, and the minimum bounding box of each merged cluster is calculated; the non-equilibrium region is constructed using the physical three-dimensional spatial range defined by the minimum bounding box of the minimum bounding box.

[0062] Specifically, the optimized heat flow path is used to perform boundary correction processing on the non-equilibrium region to obtain the heat transfer difference, including: The optimized heat flow path is mapped to the non-equilibrium region, and the local fine-tuning coefficient is calculated. Generate a boundary correction matrix based on the local fine-tuning coefficients; The boundary correction matrix is ​​input into the fluid dynamics solver to update the nodes in the non-equilibrium region, and the convective heat transfer coefficient change field is recalculated. The heat transfer difference is obtained by calculating the difference in heat dissipation efficiency in the non-equilibrium region based on the change field of the convective heat transfer coefficient.

[0063] In one implementation, this embodiment maps the optimized heat flow path to the non-equilibrium region and calculates local fine-tuning coefficients. This embodiment extracts spatial vector segments through which the optimized heat flow path passes the internal boundary of the non-equilibrium region; calculates the spatial angle between the direction vector of the spatial vector segment and the surface normal vector of the geometric entity interface where the anomalous node is located; extracts the cosine value of this spatial angle and multiplies it by the transient energy fluctuation rate value corresponding to the coordinates of the anomalous node to calculate a dimensionless coefficient quantifying the turbulent impact intensity, thus generating the local fine-tuning coefficients.

[0064] It should be noted that this embodiment generates the boundary correction matrix based on the local fine-tuning coefficients. This embodiment extracts the standard wall function parameters assigned to the abnormal node coordinates in the original fluid dynamics solver; multiplies the standard wall function parameters by the local fine-tuning coefficients to obtain the overwritten convective heat transfer multiplier parameters and wall roughness penalty parameters; performs node addressing matching on the overwritten parameters corresponding to all abnormal node coordinates within the non-equilibrium region and encapsulates them into a matrix data structure to generate the dedicated boundary correction matrix.

[0065] In one implementation, this embodiment inputs the boundary correction matrix into the fluid dynamics solver to update the nodes in the non-equilibrium region, recalculating the convective heat transfer coefficient variation field. This embodiment loads the boundary correction matrix as a local boundary overwrite file into the fluid dynamics solver's runtime memory, forcing the solver to perform local iterative recalculation of the fluid-structure interaction boundary layer only for mesh nodes located within the non-equilibrium region. The latest calculated value of the convective heat transfer coefficient for each target node after local recalculation convergence is extracted, and this value is subtracted from the historical value of the convective heat transfer coefficient corresponding to that node before local recalculation to calculate the heat transfer coefficient deviation value for each node. The heat transfer coefficient deviation values ​​are combined according to the mesh topology to generate the convective heat transfer coefficient variation field.

[0066] It should be noted that this embodiment calculates the heat transfer difference by measuring the difference in heat dissipation efficiency within the non-equilibrium region based on the convective heat transfer coefficient variation field. This embodiment extracts the heat transfer coefficient deviation value of each target node in the convective heat transfer coefficient variation field, extracts the node surface area value of the interface where the target node is located, and the absolute value of the temperature difference between the fluid and solid interfaces at that location; it then multiplies the heat transfer coefficient deviation value, the node surface area value, and the absolute value of the temperature difference to obtain the scalar value of the local heat flux change for a single target node; it then iterates through all target nodes within the non-equilibrium region, summing all the local heat flux change scalar values ​​to obtain the total total heat dissipation efficiency difference, which is output as the heat transfer difference.

[0067] For example, the system extracts the three-dimensional wind speed component time series of each grid point in the window edge region, calculates the kinetic energy variance of the fluid micro-particle by summing the arithmetic mean and the squared difference, and determines the transient energy fluctuation rate to be 0.18. The system determines that 0.18 is greater than the preset fluctuation threshold of 0.15 based on the statistical calibration of laminar flow wind tunnel experiments. The system then extracts the three-dimensional coordinates of the grid points exceeding the threshold, performs density clustering, determines the neighborhood radius using the second derivative extremum point, determines the minimum sample size using the numerical value 4, and calculates with the bounding box algorithm to define the spatial range of the window edge region as a non-equilibrium region. Subsequently, the system extracts the spatial segment of the optimized heat flow path in the non-equilibrium region, calculates the local fine-tuning coefficient by combining the cosine value of the vector angle and the fluctuation rate, and multiplies it by the original standard wall function parameters to generate a boundary correction matrix containing overriding parameters. After inputting the boundary correction matrix into the solver for local mesh node updates and recalculation, the convective heat transfer coefficient in the original window edge region was updated from 8.5 Kelvin per square meter to 12.4 Kelvin per square meter. The difference between the two is 3.9 Kelvin per square meter, forming a convective heat transfer coefficient variation field. Finally, by combining the nodal surface area with the absolute value of the fluid-solid temperature difference through multiplication and summation, the actual heat dissipation efficiency deviation in this non-equilibrium region was quantified as 35 Kelvin, which was identified as the extracted heat transfer difference.

[0068] In step S16, the ambient temperature and humidity are obtained, and the heat transfer difference and the ambient temperature and humidity are input into a preset heat and humidity coupling equation for solution to obtain the maximum fluctuation amplitude. If the maximum fluctuation amplitude is less than the preset adjustment threshold, the optimized heat flow path is topologically deduced using a pre-trained graph neural network to obtain an interaction model.

[0069] The process involves acquiring ambient temperature and humidity, inputting the heat transfer difference and the ambient temperature and humidity into a preset thermal-humidity coupling equation for solution, and obtaining the maximum fluctuation amplitude, including: Obtain ambient temperature and humidity; Substitute the heat transfer difference and the ambient temperature and humidity into a preset thermal-humidity coupling equation and perform iterative solution to obtain the dynamic thermal-humidity response value; The hourly heating and cooling load numerical sequence is derived using the dynamic thermal and humidity response values. The fluctuation sequence is obtained by performing a difference operation on the values ​​of adjacent time nodes in the hourly heating and cooling load numerical sequence, and the extreme value of the fluctuation sequence is extracted as the maximum fluctuation amplitude.

[0070] In one implementation, this embodiment acquires ambient temperature and humidity. Specifically, it collects real-time dry-bulb temperature and relative humidity values ​​through a network of temperature and humidity sensors deployed inside and outside the target building area. This embodiment substitutes the heat transfer difference and the ambient temperature and humidity into a preset thermal-humidity coupling equation and performs iterative solving to obtain the dynamic thermal-humidity response value. It should be noted that the preset thermal-humidity coupling equation is a set of partial differential equations based on the principles of energy conservation and water mass conservation. The specific operational logic of the energy conservation equation is shown in the following formula:

[0071] in, It represents the sum of sensible and latent heat energy per unit volume of a node. Its initial value is calculated by combining real-time data collected by environmental sensors deployed in the building with the density of building materials, and the dimension is joules per cubic meter. Represents spatial gradient and divergence operations; The time step is determined based on the Courant-Friedrich-Levi stability condition in numerical analysis. Specifically, the method is to extract the local maximum velocity value in the velocity vector field, calculate the quotient of the two by combining the minimum characteristic length of the numerical calculation grid space, and multiply it by a preset safety convergence coefficient to ensure the convergence of the iterative calculation. The thermal conductivity of building materials is represented by a standard building material property parameter database that contains pre-stored physical measurement values ​​of different materials. This represents the node temperature, whose initial value originates from the temperature scalar field obtained in S11 using a computational fluid dynamics algorithm. and These represent air density and specific heat capacity, respectively. Their values ​​are determined by performing a meteorological property lookup method in a standard meteorological property parameter database based on the real-time dry-bulb temperature and standard atmospheric pressure fed back by current environmental sensors. This represents the airflow velocity vector, the value of which is obtained from the latest velocity vector field result calculated in the previous steps; This represents the latent heat of vaporization of water vapor, and its value is determined by retrieving the standard saturated water vapor property table based on the current node temperature. This represents water vapor permeability, and its value is obtained by retrieving the standard building material property parameter database and matching the material label of the corresponding building component. It represents the partial pressure of water vapor, and its value is derived by multiplying the relative humidity collected by the sensor with the saturated vapor pressure at the current temperature.

[0072] It should be noted that the preset safe convergence coefficient was determined through multiple trial calculations using typical annual hourly meteorological data of the target building's location. Specifically, the safe convergence coefficients were set to 0.2, 0.4, 0.6, 0.8, and 1.0, respectively, and the thermo-humid coupling equation was iteratively solved. The calculation results were observed to ensure stable convergence without numerical oscillations. The smallest coefficient value that enables stable convergence of the iteration was selected as the preset safe convergence coefficient, which is typically between 0.3 and 0.7.

[0073] In this embodiment, the implicit finite difference algorithm is used to discretize the partial differential equation system in the time domain, and the result is input into a numerical solver to perform iterative calculation. The output is a numerical solution that reflects the evolution of the spatial thermal and humidity state with time step, which is determined as the dynamic thermal and humidity response value.

[0074] In one implementation, this embodiment uses the dynamic thermal and humidity response values ​​to deduce an hourly sequence of cooling and heating load values. This embodiment performs integral and cumulative calculations on the dynamic thermal and humidity response values ​​within a future operating cycle according to a preset time integration window, deduces the required cooling or heating compensation values ​​at each time point, and arranges the cooling or heating values ​​in chronological order into an array to generate the hourly sequence of cooling and heating load values. It should be noted that the preset time integration window is determined by reading the data sampling frequency calibration from the control panel of the HVAC system in the target area.

[0075] It should be noted that in this embodiment, the fluctuation sequence is obtained by performing a difference operation on the values ​​of adjacent time nodes in the hourly heating and cooling load numerical sequence. Specifically, in the hourly heating and cooling load numerical sequence, the heating and cooling load value of the previous time node is subtracted from the heating and cooling load value of the next time node to calculate the load step change between adjacent time nodes; all load step changes generated in chronological order are used to form the fluctuation sequence. In this embodiment, an extreme value operation is performed on the fluctuation sequence, and the maximum absolute value in the fluctuation sequence is extracted and determined as the maximum fluctuation amplitude.

[0076] It is worth noting that in this embodiment, the conditional judgment action, if the maximum fluctuation amplitude is less than a preset adjustment threshold, triggers the subsequent multidimensional vector field generation and topology deduction process. The preset adjustment threshold is determined by reading the instruction manual of the target building's air conditioning terminal equipment, extracting its rated stable adjustment power upper limit value, and directly assigning it.

[0077] The optimized heat flow path is topologically derived using a pre-trained graph neural network to obtain an interaction model, including: The optimized heat flow path is mapped to a preset three-dimensional mesh space using an inverse distance weighted interpolation algorithm to construct a multi-dimensional vector field; Thermal gradient calculation is performed on the multidimensional vector field to extract the interaction weight values, and the interaction weight values ​​are arranged in order to construct a thermal-humid coupling feature matrix; A topology diagram is constructed using the aforementioned thermal-humidity coupling feature matrix; The topology graph is input into a pre-trained graph neural network, and the evolution path sequence is derived using a dynamic path optimization algorithm. The evolutionary path sequences are aggregated to obtain an interaction model.

[0078] In one implementation, this embodiment uses an inverse distance weighted interpolation algorithm to map the optimized heat flow path to a preset three-dimensional mesh space, constructing a multidimensional vector field. This embodiment divides the physical space into the preset three-dimensional mesh space with a fixed discrete resolution.

[0079] It should be noted that the preset discrete resolution of the three-dimensional mesh space is the volume cell size determined by performing a spatial mesh independence sensitivity test on the target control area. The criterion for the sensitivity test is that when the absolute value of the global average temperature difference calculated by two adjacent mesh subdivision operations is less than 0.1 degrees Celsius, the volume cell size of the current level is determined as the discrete resolution. In this embodiment, discrete data points on the optimized heat flow path are extracted as source points. For any unassigned mesh center point in the preset three-dimensional mesh space, known source points in its neighborhood are searched, and the spatial straight-line distance from the mesh center point to each known source point is calculated. The reciprocal of the spatial straight-line distance is used as the interpolation weight to perform a weighted average calculation of the heat flow direction vector and humidity diffusion scalar of the known source points, and this average is assigned to the mesh center point, thereby constructing a continuous multi-dimensional vector field containing the heat flow direction and the humid air diffusion trend between discrete points.

[0080] In one implementation, this embodiment performs thermal gradient calculations on the multidimensional vector field to extract interaction weight values, and then arranges these interaction weight values ​​in an ordered manner to construct a thermal-humidity coupling feature matrix. This embodiment calculates the spatial partial derivatives of temperature and humidity between adjacent grid nodes in the three-dimensional direction of the multidimensional vector field, extracts the absolute values ​​of these spatial partial derivatives, and performs global maximum-minimum normalization on them, mapping the values ​​to a dimensionless interval of zero to one, and determines these as the interaction weight values. The interaction weight values ​​of all grid nodes in the multidimensional vector field are arranged into a two-dimensional array according to their spatial three-dimensional coordinate index positions to generate the thermal-humidity coupling feature matrix, which digitally represents the intrinsic correlation strength of energy and mass transfer in space.

[0081] It should be noted that this embodiment utilizes the aforementioned thermal-humidity coupling feature matrix to construct a topology graph. This embodiment extracts the coordinates of thermal-humidity state points in the preset three-dimensional grid space to establish nodes in the graph, and establishes edges between adjacent nodes that are logically connected by the direction of heat flow evolution. Specifically, the logic for establishing edges is as follows: the dot product of the multidimensional vector field direction vectors between adjacent grid nodes is calculated. If the dot product result is greater than zero, it is determined that there is a positive fluid convection transport path between the two nodes, and a directed edge is established between the two nodes. Based on the coordinate index corresponding to the node, the corresponding interaction weight value is extracted from the thermal-humidity coupling feature matrix and assigned to the corresponding edge as an edge weight scalar, generating the topology graph reflecting the physical morphology of airflow organization.

[0082] In one implementation, this embodiment inputs the topology graph into a pre-trained graph neural network and uses a dynamic path optimization algorithm to deduce the evolutionary path sequence. This embodiment inputs the node feature matrix and edge weight adjacency matrix representing the topology graph into the pre-trained graph neural network; through the message passing mechanism of the graph convolutional layer and the weighted aggregation operation of neighbor node features, it updates and generates a node hidden layer representation containing local flow field spatiotemporal features; in the output layer of the graph neural network, it calls a weight variant of the Dijkstra's shortest path algorithm, using the reciprocal of the updated node interaction weight values ​​as the path impedance, to search for the grid connection sequence with the minimum total impedance between the heat source node and the exhaust vent node, and outputs it as the evolutionary path sequence.

[0083] It is worth noting that the pre-trained graph neural network is configured with three consecutive graph convolutional message-passing layers. Its input node feature vectors are multi-dimensional and one-dimensional arrays formed by sequentially concatenating temperature values, humidity values, spatial coordinates, and wind speed scalars along the feature dimensions. During the model pre-training phase, a large dataset of historical high-precision airflow thermal coupling simulation steady-state airflow topology structures and corresponding real airflow trajectories are acquired as training samples. The predicted path output by the network and the real trajectory are substituted into the cross-entropy loss function to calculate the reconstruction error. An adaptive moment estimation optimization algorithm combined with a backpropagation mechanism is used to iteratively update the connection weights of the network layers, and an exponential decay strategy for the learning rate is employed to prevent the model from getting trapped in local optima.

[0084] In one implementation, this embodiment aggregates the evolution path sequences to obtain an interaction model. This embodiment extracts multiple evolution path sequences originating from different starting nodes from the deduced data and performs spatial graphic overlay assembly in the same virtual three-dimensional coordinate system to generate a panoramic map that fully reproduces the process of heat flow interfering with airflow guidance and reshaping local heat distribution. This map is then stored and output as the interaction model.

[0085] For example, the system extracts the absolute value of the cooling load difference of 8 kW between 5 PM and 6 PM as the maximum fluctuation amplitude. With a threshold of 10 kW for the comparison device, the system determines that 8 kW is less than the upper limit of the allowable fluctuation threshold. Subsequently, the system uses a reciprocal weighted algorithm to map the optimized heat flow path onto a discrete three-dimensional grid space divided into 0.125 cubic meters to construct a multi-dimensional vector field. After thermal gradient partial derivative calculation and maximum-minimum normalization processing, it is assembled into a thermal-humidity coupling feature matrix. The weight values ​​of this matrix are used to assign values ​​to the edges to construct a topological structure graph reflecting the airflow organization. This graph is input into a pre-trained graph neural network configured with three continuous graph convolutional layers. Through local feature aggregation and dynamic path optimization using the inverse impedance feature to find the path with the minimum total impedance, the system outputs an evolution path sequence describing the trajectory of the humid and hot airflow spreading along the ceiling to the return air vent. The system aggregates and superimposes multiple evolution path sequences from different starting points to assemble and output a dynamic interaction model.

[0086] It should be further explained that if the influence weight of airflow is not greater than the preset weight threshold, the interaction inference step using deep convolutional neural network is skipped, the initial heat flow path is directly used as the optimized heat flow path, and subsequent steps are continued; if the transient energy fluctuation rate is not greater than the preset fluctuation threshold, the boundary correction process of the non-equilibrium region is not performed, the heat transfer difference is set to zero, and step S16 is continued; if the maximum fluctuation amplitude is not less than the preset adjustment threshold, the current optimized heat flow path is directly output as the final interaction model, and the topology inference of graph neural network is no longer performed.

[0087] In summary, this invention innovatively integrates computational fluid dynamics simulation with a pre-trained deep convolutional neural network to interactively extrapolate the directional velocity component and the initial heat flow path to predict the heat distribution. This overcomes the feature extraction bottleneck of traditional linear models when dealing with strongly disturbed airflows, achieving high-precision identification of nonlinear coupling deviations in heat flow under complex urban microclimates. By dynamically adjusting controllable building components based on deviation feedback to reset the fluid calculation boundary, and constructing non-equilibrium regions for transient energy fluctuations exceeding thresholds to perform local boundary correction and heat transfer recalculation, the passive static environment measurement is transformed into an active physical boundary closed-loop optimization. This achieves dynamic adaptive feedback of building space morphology to changes in airflow and heat, effectively eliminating the risks of local heat accumulation and turbulent oscillations. Furthermore, by leveraging the extreme value compliance verification of the thermal-humidity coupling equation and the topological extrapolation of the optimized heat flow path using a graph neural network, the isolated single-point energy consumption assessment is upgraded to a panoramic dynamic map of the interaction between airflow and heat. This enables full-time, high-fidelity simulation and adaptive collaborative control of the microclimate environment of green buildings, effectively solving the industry problem of the mutual constraints between heat flow and airflow guidance in complex urban environments, which is difficult to simulate accurately. It provides a scientific, reliable, and intelligent decision-making basis for refined air volume distribution and emission reduction and accumulation control, and significantly improves the overall ecological regulation efficiency and energy-saving performance of high-density urban building clusters.

[0088] Reference Figure 2 The second embodiment of the present invention provides a green building microclimate regulation, emission reduction and sequestration system based on cities, comprising: The data processing module is used to acquire point cloud data of the building and surrounding environmental parameters, and use computational fluid dynamics algorithms to solve the point cloud data and environmental parameters to obtain the initial heat flow path; The convection calculation module is used to perform convection heat transfer calculation based on the initial heat flow path to obtain the directional velocity component and the airflow influence weight; The feature inference module is used to perform interactive inference on the directional velocity component and the initial heat flow path to obtain a predicted heat distribution if the influence weight of the airflow is greater than a preset weight threshold, and compare the predicted heat distribution with the initial heat flow path to obtain a deviation value matrix. The boundary calculation module is used to perform layout optimization based on the deviation value matrix and preset component constraints to obtain adjusted layout parameters, and to use the adjusted layout parameters to perform boundary calculation to obtain optimized heat flow path and velocity vector field. The energy fluctuation module is used to determine the transient energy fluctuation rate based on the velocity vector field, and when the transient energy fluctuation rate is greater than a preset fluctuation threshold, to construct a non-equilibrium region using the velocity vector field, and to perform boundary correction processing on the non-equilibrium region using the optimized heat flow path to obtain the heat transfer difference; The interactive modeling module is used to acquire the ambient temperature and humidity, and input the heat transfer difference and the ambient temperature and humidity into a preset heat and humidity coupling equation for solving to obtain the maximum fluctuation amplitude. If the maximum fluctuation amplitude is less than the preset adjustment threshold, the pre-trained graph neural network is used to perform topological deduction on the optimized heat flow path to obtain the interaction model.

[0089] It should be noted that the green building microclimate regulation, emission reduction and carbon sequestration enhancement system based on urban areas provided in this embodiment of the invention is used to execute all the process steps of the green building microclimate regulation, emission reduction and carbon sequestration enhancement method based on urban areas in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0090] It should be noted that the system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the system embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0091] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that 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 for those skilled in the art.

Claims

1. A city-based green building micro-climate regulation emission reduction and carbon sink increase method, characterized in that, include: The point cloud data of the building and the surrounding environmental parameters are acquired, and the initial heat flow path is obtained by solving the point cloud data and the environmental parameters using computational fluid dynamics algorithms. Based on the initial heat flow path, convective heat transfer calculations are performed to obtain the directional velocity component and the airflow influence weight; If the airflow influence weight is greater than the preset weight threshold, a pre-trained deep convolutional neural network is used to interactively deduce the directional velocity component and the initial heat flow path to obtain the predicted heat distribution, and the predicted heat distribution is compared with the initial heat flow path to obtain the deviation value matrix. The layout is optimized based on the deviation matrix and the preset component constraints to obtain the adjusted layout parameters, and the boundary calculation is performed using the adjusted layout parameters to obtain the optimized heat flow path and velocity vector field. The transient energy fluctuation rate is determined based on the velocity vector field, and when the transient energy fluctuation rate is greater than a preset fluctuation threshold, a non-equilibrium region is constructed using the velocity vector field. The boundary correction process of the non-equilibrium region is performed using the optimized heat flow path to obtain the heat transfer difference. The ambient temperature and humidity are obtained, and the heat transfer difference and the ambient temperature and humidity are input into a preset heat and humidity coupling equation for solution to obtain the maximum fluctuation amplitude. If the maximum fluctuation amplitude is less than the preset adjustment threshold, the optimized heat flow path is topologically deduced using a pre-trained graph neural network to obtain an interaction model.

2. The urban-based green building micro-climate regulation carbon emission reduction and sink enhancement method according to claim 1, characterized in that, The process of acquiring point cloud data of the building and surrounding environmental parameters, and using computational fluid dynamics algorithms to solve for the point cloud data and environmental parameters to obtain the initial heat flow path, includes: Acquire point cloud data and extract surface features from the point cloud data to obtain material layout properties; Construct a building entity model based on the aforementioned material layout properties; The building entity model is discretized based on environmental parameters to generate a numerical computation grid space; The velocity vector field and temperature scalar field are obtained by solving the numerical computation grid space using computational fluid dynamics algorithms. The instantaneous heat flux distribution is calculated based on the velocity vector field and the temperature scalar field. The initial heat flux path is obtained by extracting the spatial evolution trajectory of the instantaneous heat flux distribution.

3. The urban based green building micro-climate regulation carbon emission reduction and sink enhancement method as claimed in claim 1 wherein, The step of performing convective heat transfer calculations based on the initial heat flow path to obtain the directional velocity component and airflow influence weights includes: Node parameters are extracted based on the initial heat flow path, and scalar and vector feature separation is performed on the node parameters to obtain the airflow velocity field and airflow direction field; The directional velocity component is obtained by spatial vector fusion of the airflow velocity field and the airflow direction field. By combining the preset air density and the preset specific heat capacity, the convective heat transfer capacity of the grid nodes is calculated using the directional velocity component to obtain the convective heat transfer intensity matrix. Spatial correlation analysis was performed on the convective heat transfer intensity matrix to obtain the airflow guidance coupling coefficient; The airflow influence weights are obtained by solving the airflow guidance coupling coefficient using a multiple linear regression algorithm.

4. The urban-based green building micro-climate regulation carbon emission reduction and sink enhancement method according to claim 1, wherein, The method involves using a pre-trained deep convolutional neural network to interactively extrapolate the directional velocity component and the initial heat flow path to obtain a predicted heat distribution, and then comparing the predicted heat distribution with the initial heat flow path to obtain a deviation matrix, including: The directional velocity component and the initial heat flow path are mapped to a preset three-dimensional mesh space; Within the three-dimensional mesh space, the directional velocity component and the initial heat flow path are tensor-concatenated to generate a spatial feature vector; The spatial feature vector is input into a pre-trained deep convolutional neural network, and deep interaction feature matrix is ​​extracted using multi-layer convolution operations. The predicted heat distribution is obtained by decoding and deducing the deep interaction feature matrix using a pre-trained decoder. The predicted heat distribution is compared with the initial heat flow path point by point to obtain the deviation matrix by calculating the difference.

5. The method for green building microclimate regulation, emission reduction, and carbon sequestration enhancement based on urban areas according to claim 1, characterized in that, The step of optimizing the layout based on the deviation matrix and preset component constraints to obtain adjusted layout parameters, and then using the adjusted layout parameters to perform boundary calculations to obtain optimized heat flow paths and velocity vector fields, includes: Based on the deviation value matrix, a correction factor is assigned to the preset three-dimensional grid space to generate a spatial correction factor matrix; The spatial correction factor matrix is ​​used to generate shape adjustment instructions within the preset component constraints, and the adjusted layout parameters are generated according to the shape adjustment instructions. The preset computational boundary conditions are reset using the adjusted layout parameters to construct an updated thermodynamic conduction model; The updated thermodynamic conduction model is input into the fluid dynamics solver for global solution to obtain the optimized heat flow path and velocity vector field.

6. The method for green building microclimate regulation, emission reduction, and carbon sequestration enhancement based on urban areas according to claim 1, characterized in that, The step of determining the transient energy fluctuation rate based on the velocity vector field, and constructing a non-equilibrium region using the velocity vector field when the transient energy fluctuation rate is greater than a preset fluctuation threshold, includes: Extract the three-dimensional wind speed components of each node in the velocity vector field; The transient energy fluctuation rate is obtained by calculating the kinetic energy variance using the three-dimensional wind speed components. Determine whether the transient energy fluctuation rate is greater than a preset fluctuation threshold; If the transient energy fluctuation rate is greater than the preset fluctuation threshold, the corresponding abnormal node coordinates are extracted from the velocity vector field; The coordinates of the abnormal nodes are used to define the spatial range and construct a non-equilibrium region.

7. The method for green building microclimate regulation, emission reduction, and carbon sequestration enhancement based on urban areas according to claim 1, characterized in that, The step of using the optimized heat flow path to perform boundary correction processing on the non-equilibrium region to obtain the heat transfer difference includes: The optimized heat flow path is mapped to the non-equilibrium region, and the local fine-tuning coefficient is calculated. Generate a boundary correction matrix based on the local fine-tuning coefficients; The boundary correction matrix is ​​input into the fluid dynamics solver to update the nodes in the non-equilibrium region, and the convective heat transfer coefficient change field is recalculated. The heat transfer difference is obtained by calculating the difference in heat dissipation efficiency in the non-equilibrium region based on the change field of the convective heat transfer coefficient.

8. The method for green building microclimate regulation, emission reduction, and carbon sequestration enhancement based on urban areas according to claim 1, characterized in that, The process of acquiring ambient temperature and humidity, and inputting the heat transfer difference and the ambient temperature and humidity into a preset thermal-humidity coupling equation for solution to obtain the maximum fluctuation amplitude includes: Obtain ambient temperature and humidity; Substitute the heat transfer difference and the ambient temperature and humidity into a preset thermal-humidity coupling equation and perform iterative solution to obtain the dynamic thermal-humidity response value; The hourly heating and cooling load numerical sequence is derived using the dynamic thermal and humidity response values. The fluctuation sequence is obtained by performing a difference operation on the values ​​of adjacent time nodes in the hourly heating and cooling load numerical sequence, and the extreme value of the fluctuation sequence is extracted as the maximum fluctuation amplitude.

9. The method for green building microclimate regulation, emission reduction, and carbon sequestration enhancement based on urban areas according to claim 1, characterized in that, The method of using a pre-trained graph neural network to perform topology deduction on the optimized heat flow path to obtain an interaction model includes: The optimized heat flow path is mapped to a preset three-dimensional mesh space using an inverse distance weighted interpolation algorithm to construct a multi-dimensional vector field; Thermal gradient calculation is performed on the multidimensional vector field to extract the interaction weight values, and the interaction weight values ​​are arranged in order to construct a thermal-humid coupling feature matrix; A topology diagram is constructed using the aforementioned thermal-humidity coupling feature matrix; The topology graph is input into a pre-trained graph neural network, and the evolution path sequence is derived using a dynamic path optimization algorithm. The evolutionary path sequences are aggregated to obtain an interaction model.

10. A green building microclimate regulation, emission reduction, and carbon sequestration enhancement system based on urban areas, characterized in that: include: The data processing module is used to acquire point cloud data of the building and surrounding environmental parameters, and use computational fluid dynamics algorithms to solve the point cloud data and environmental parameters to obtain the initial heat flow path; The convection calculation module is used to perform convection heat transfer calculation based on the initial heat flow path to obtain the directional velocity component and the airflow influence weight; The feature inference module is used to perform interactive inference on the directional velocity component and the initial heat flow path to obtain a predicted heat distribution if the influence weight of the airflow is greater than a preset weight threshold, and compare the predicted heat distribution with the initial heat flow path to obtain a deviation value matrix. The boundary calculation module is used to perform layout optimization based on the deviation value matrix and preset component constraints to obtain adjusted layout parameters, and to use the adjusted layout parameters to perform boundary calculation to obtain optimized heat flow path and velocity vector field. The energy fluctuation module is used to determine the transient energy fluctuation rate based on the velocity vector field, and when the transient energy fluctuation rate is greater than a preset fluctuation threshold, to construct a non-equilibrium region using the velocity vector field, and to perform boundary correction processing on the non-equilibrium region using the optimized heat flow path to obtain the heat transfer difference; The interactive modeling module is used to acquire the ambient temperature and humidity, and input the heat transfer difference and the ambient temperature and humidity into a preset heat and humidity coupling equation for solving to obtain the maximum fluctuation amplitude. If the maximum fluctuation amplitude is less than the preset adjustment threshold, the pre-trained graph neural network is used to perform topological deduction on the optimized heat flow path to obtain the interaction model.

Images