Curtain wall refined model generation method and system

Abstract

The invention discloses a curtain wall refined model generation method and system, and relates to the technical field of building model generation, and the method comprises the steps: collecting multi-source heterogeneous data of a target building, generating a curtain wall boundary line according to a building contour line, carrying out the segmentation parameter bidirectional dynamic adjustment and adaptive segmentation spline interpolation of the curtain wall boundary line, and generating a basic curtain wall basic surface; the method comprises the following steps: constructing a basic three-dimensional geometric model, generating an initial lattice surface, constructing a thermal response model by combining the basic three-dimensional geometric model, material characteristic data and thermal response data, obtaining thermal characteristic data of each region, optimizing the lattice surface of the basic three-dimensional geometric model according to the thermal characteristic data, and generating a three-dimensional geometric model. Performing automatic grouping and panel generation operation on the three-dimensional geometric model; and performing multi-modal feature fusion on the three-dimensional geometric model, the thermal feature data and the material feature data to generate a three-dimensional curtain wall refined model, and performing multi-physical field coupling verification on the three-dimensional curtain wall refined model, so that the curtain wall model generation efficiency and precision are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of building model generation, in particular to a curtain wall fine model generation method and system. BACKGROUND

[0002] A curtain wall fine model generation method and system based on curtain wall marginal line is disclosed in Chinese patent CN114781030A, which includes the following steps: generating curtain wall marginal line based on preset modeling elements; creating curtain wall point array surface based on the curtain wall marginal line; dividing array points in the curtain wall point array surface into multiple point array groups through grouping; generating curtain wall model in combination with preset adaptive panel and multiple point array groups.

[0003] A curtain wall modeling method and system based on BIM model are disclosed in Chinese patent CN112149211A, which includes the following steps: S11. Identifying curtain wall nodes through node identification software and uploading the identified curtain wall nodes to curtain wall modeling software; S12. Establishing curtain wall 3D facade grid based on curtain wall modeling software; S13. Selecting curtain wall sub-grid of unit panel in the established 3D facade grid and placing the curtain wall nodes identified in step S11 at the node position of the selected curtain wall sub-grid to generate different standard panel template models; S14. Establishing the entire curtain wall model according to the generated standard panel template models.

[0004] In the construction industry, curtain wall as the outer envelope structure of the building, the degree of its design and construction refinement is crucial to the aesthetics, performance and safety of the building. Traditional curtain wall model generation methods often have problems such as insufficient precision and inability to fully reflect various characteristics of the curtain wall. For example, in the construction of geometric shapes, it is difficult to accurately simulate the subtle changes of complex curved curtain walls; in terms of material and thermal characteristics, there is a lack of effective fusion means, resulting in models that cannot meet the needs of performance analysis, construction guidance and post-maintenance of curtain walls in actual engineering. With the development of the construction industry, the demand for curtain wall fine models is increasing, and there is an urgent need for a method that can consider multiple factors and generate high-precision curtain wall fine models with rich information. SUMMARY

[0005] To solve the above technical problems, the purpose of the present application is to provide a curtain wall fine model generation method, which includes the following steps: Step s1: Collecting multi-source heterogeneous data of target building and user intervention instructions, and performing data preprocessing on the multi-source heterogeneous data, the multi-source heterogeneous data including high-precision point cloud data, building line contour, environmental data, material property data, thermal response data, wind load data and temperature stress data; Step s2: generating curtain wall marginal lines according to the building contour line, performing two-way dynamic adjustment of the segmentation parameters of the curtain wall marginal lines, and performing adaptive piecewise spline interpolation to generate smooth curves, and generating a basic curtain wall base surface based on the smooth curves; Step s3: constructing a basic three-dimensional geometric model according to the basic curtain wall base surface, generating an initial point array surface in combination with the segmentation points of the curtain wall marginal lines, constructing a thermal response model in combination with the basic three-dimensional geometric model, material characteristic data, and thermal response data, obtaining thermal characteristic data of each region, performing point array surface optimization of the basic three-dimensional geometric model according to the thermal characteristic data, generating a three-dimensional geometric model, and performing automatic grouping and panel generation operations on the three-dimensional geometric model; Step s4: performing multi-modal feature fusion on the three-dimensional geometric model, thermal characteristic data, and material characteristic data to generate a three-dimensional curtain wall refined model, and performing multi-physical field coupling verification on the three-dimensional curtain wall refined model.

[0006] Further, the process of generating a curtain wall marginal line according to the building contour line, performing two-way dynamic adjustment of the segmentation parameters of the curtain wall marginal line, and performing adaptive piecewise spline interpolation to generate smooth curves, and generating a basic curtain wall base surface based on the smooth curves includes: According to the building contour line, a parameterized design method is used to select target modeling elements to generate curtain wall marginal lines. Parameterized design is a powerful design method that allows designers to control the geometric shape of the model by adjusting parameters. In generating curtain wall marginal lines, spline curves, geometric shapes, etc. can be selected as target modeling elements. For example, in designing a curtain wall with a smooth curve shape, by setting control points and adjusting parameters, spline curves can easily follow the unique contour of the building, ensuring that the curtain wall marginal line perfectly matches the appearance of the building. For some regular-shaped curtain wall parts, such as rectangles, triangles, etc. geometric shapes are more suitable. They can quickly build the basic framework of the curtain wall marginal line by simply defining parameters such as side length, angle, etc. According to the pre-processed wind load data and temperature stress data, as well as user intervention instructions, the segmentation parameters of the curtain wall marginal line are adjusted dynamically in two directions. According to the segmentation parameters after two-way dynamic adjustment, several segmentation points are marked on the curtain wall marginal line. Key control points are selected on the curtain wall marginal line, and adaptive piecewise spline interpolation is performed on the curtain wall marginal line according to the key control points to generate smooth curves. The lofting operation is performed on several smooth curves, i.e. connecting these curves along a certain path to generate a continuous basic curtain wall base surface.

[0007] Further, the process of marking several segmentation points on the curtain wall marginal line according to the segmentation parameters includes: Get the dynamic segmentation spacing corresponding to each initial node. Assume that the dynamic segmentation spacing corresponding to the node Pi is di. This spacing is the result of adjustment based on the specific situation of each node relative to the benchmark segmentation spacing. It reflects the actual segmentation density required at the node. Start marking from the starting node, select an endpoint of the curtain wall edge line as the starting point, and find the initial node closest to the endpoint as the first node to be processed, assuming it is P1. Take P1 as the starting point, and measure the distance d1 along the direction of the edge line on the curtain wall edge line according to its corresponding dynamic segmentation spacing d1, and mark a new point Q1. This point is the first segmentation point determined according to the dynamic segmentation spacing of P1, and then mark other segmentation points in turn. For the next initial node P2, take the already marked point Q1 as the starting point, measure the distance d2 along the direction of the edge line, and mark the new segmentation point Q2. In this way, for each initial node Pi, the segmentation point marked by the previous one is marked. Starting from the edge, measure the distance di along the edge and mark the corresponding segmentation point Qi. If the boundary or endpoint of the curtain wall edge is encountered during the marking process, the marking method is adjusted. If the remaining edge length is less than the dynamic segmentation spacing di of the current node, the remaining length is used as the segmentation spacing, and the last segmentation point is marked so that it is exactly at the endpoint of the edge. If, during the marking process, it is found that the segmentation points corresponding to two adjacent initial nodes may overlap or the spacing is too small (less than the preset minimum spacing threshold), the position of the segmentation points is fine-tuned, and the segmentation spacing of one of the nodes is appropriately increased to ensure that the segmentation points are reasonably spaced and can accurately reflect the characteristics and design requirements of the curtain wall edge.

[0008] Furthermore, the process of bidirectionally dynamically adjusting the segmentation parameters of the curtain wall edge line according to the pre-processed wind load data and temperature stress data and user intervention instructions includes: Preset the benchmark division spacing, structural response weight, and curvature weight, and evenly generate the initial point set along the curtain wall edge line according to the benchmark division spacing. , perform differential operations on each initial node in the initial point set to obtain the curvature value k of each initial node, map the wind load data and temperature stress data to the initial point set for multi-field coupling analysis, obtain the total stress coefficient of each initial node, and obtain the dynamic segmentation spacing corresponding to each initial node based on the total stress coefficient and curvature value of each initial node and the structural response weight and curvature weight; The calculation formula for obtaining the dynamic segmentation distance of each initial node is: ; ; ; ; ; wherein, denotes the dynamic partition interval of the initial node i, denotes the reference partition interval, denotes the structural response weight, denotes the curvature weight, denotes the total stress coefficient of the initial node i, denotes the curvature value of the initial node i, denotes the wind load stress coefficient, denotes the temperature stress coefficient, denotes the wind load sensitivity coefficient, denotes the temperature stress sensitivity coefficient, denotes the wind vibration coefficient, denotes the wind pressure height variation coefficient, denotes the building shape coefficient, denotes the material elastic modulus, denotes the material linear expansion coefficient, denotes the temperature difference, denotes the local annual temperature extreme value; According to the user intervention instruction, the partition interval fixed threshold interval corresponding to each initial node is obtained, it is judged whether the dynamic partition interval corresponding to each initial node is located in the corresponding partition interval fixed threshold interval, if yes, the dynamic partition interval of the initial node is marked as the best partition interval, if not, the best partition interval of the initial node is set according to the partition interval fixed threshold interval, the maximum partition interval in the partition interval fixed threshold interval is selected as the best partition interval, and the partition parameters are constructed according to the best partition interval corresponding to each initial node.

[0009] Further, the process of selecting the key control points on the curtain wall marginal line, and performing adaptive segmented spline interpolation on the curtain wall marginal line according to the key control points to generate a smooth curve includes: The curvature value and the total stress coefficient of each partition point on the curtain wall marginal line are weighted and averaged to obtain the key coefficient of each partition point, a key coefficient threshold is preset, the key coefficient of each partition point is compared with the key coefficient threshold, the partition point with the key coefficient greater than the key coefficient threshold is marked as a key control point, the curvature variation analysis and the total stress variation analysis are performed on the adjacent key control points, and the curvature variation coefficient and the total stress variation coefficient between the adjacent key control points are obtained. The preset curvature change coefficient and the total stress change coefficient correspond to threshold values, whether the curvature change coefficient and the total stress change coefficient between adjacent key control points are less than the corresponding threshold values are judged, if both are less than, a smooth curve is generated between the adjacent key control points by using a cubic spline interpolation method, if not both are less than, a smooth curve is generated between the adjacent key control points by using a B-spline interpolation method; The smooth curve between each adjacent key control point is analyzed and iteratively optimized, and a plurality of discrete points on the curtain wall marginal line between the adjacent key control points are extracted The minimum distance of the discrete points to the smooth curve between the adjacent key control points is calculated A maximum error threshold is set, if is greater than the maximum error threshold, a new node and a control point are added by using a node insertion algorithm (such as Cox-deBoor algorithm), and re-interpolation is performed, and when all discrete point errors are less than or equal to the maximum error threshold, or the iteration number reaches a preset upper limit, the optimization is stopped.

[0010] The curvature change analysis and the total stress change analysis are performed on the adjacent key control points, and the calculation formula of the curvature change coefficient and the total stress change coefficient between the adjacent key control points is obtained as follows: Let the total stress coefficients of the adjacent key control points and be and , and the curvature values be and ; ; ; Wherein, represents the curvature change coefficient, represents the total stress change coefficient.

[0011] Further, the initial point array is generated combined with the segmentation points of the curtain wall marginal line, the point array optimization of the basic three-dimensional geometric model is performed, and the process of generating the three-dimensional geometric model includes: The surface parameter space of the basic curtain wall base surface is mapped to the physical coordinate system, the basic three-dimensional geometric model is constructed, the segmentation points of the curtain wall marginal line are mapped to the basic curtain wall base surface by using the nearest point projection algorithm, the initial point array is generated, the initial point array is Delaunay triangulated, and a plurality of initial triangular meshes are generated; The thermal characteristic data includes thermal gradient, heat flux density, temperature fluctuation frequency, the thermal characteristic data of each region is used as an evaluation index, the index weight of the evaluation index is set, and the membership matrix of each region to the preset thermal sensitivity level is obtained by fuzzy comprehensive evaluation; According to the membership matrix and the index weight, the heat-sensitive grade of each region is obtained, a heat-sensitive grade threshold is preset, a region with a heat-sensitive grade greater than the heat-sensitive grade threshold is marked as a heat-sensitive region, grid density analysis is performed according to the heat characteristic data of the heat-sensitive region, the optimized grid density of the heat-sensitive region is obtained, the initial triangular meshes in the heat-sensitive region in the basic three-dimensional geometric model are re-divided according to the optimized grid density, and the three-dimensional geometric model is generated.

[0012] Further, the process of performing grid density analysis according to the heat characteristic data of the heat-sensitive region to obtain the optimized grid density of the heat-sensitive region includes: ; Among them, is the optimized grid size, represents the initial grid size, represents the thermal gradient, represents the maximum local thermal gradient, represents the heat flux density, represents the material thermal conductivity, represents the temperature fluctuation frequency (unit: Hz, such as the fluctuation caused by periodic solar radiation), represents the frequency correlation coefficient (typical value 0.1~0.3, needs to match the time step), 、 and represent the weight factor, the above formulas are to remove the dimension and take the numerical value, the formula is obtained by software simulation of a large amount of data to obtain a formula closest to the real situation, and the preset parameters and the preset threshold in the formula are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation; Further, the process of constructing a heat response model in combination with the basic three-dimensional geometric model, the material characteristic data and the heat response data to obtain the heat characteristic data of each region includes: Collecting the material characteristic data and the heat response data of each initial triangular mesh in the basic three-dimensional geometric model, performing grid optimization preprocessing on each initial triangular mesh according to the material characteristic data of each initial triangular mesh, obtaining each triangular mesh after grid optimization preprocessing, constructing a heat response model in combination with the material characteristic data and the heat response data of each triangular mesh, verifying and optimizing the heat response model, and obtaining the heat response model after verification and optimization; According to the heat response model, the heat characteristic data of each region of the basic three-dimensional geometric model is output.

[0013] Among them, the process of performing grid optimization preprocessing on each initial triangular mesh according to the material characteristic data of each initial triangular mesh includes: Adjust the mesh size of each initial triangular mesh according to the material characteristic data of each initial triangular mesh, and the calculation formula is: ; wherein, represents the adjusted mesh size, represents the mesh size before adjustment; Further, the mathematical model system required for constructing the thermal response model in combination with the material characteristic data and the thermal response data of each triangular mesh is as follows: Transient heat conduction control equation: ; wherein, is the temperature field, t is the time, is the density, is the specific heat capacity, is the thermal conductivity, is the heat source (such as solar radiation heat production).

[0014] External surface solar radiation boundary: ; wherein, is the surface absorptivity, is the solar irradiance, is the emissivity, is the Stefan-Boltzmann constant, is the external surface convective heat transfer coefficient, is the ambient temperature.

[0015] Internal surface convection boundary: ; wherein, is the internal surface convective heat transfer coefficient, is the indoor air temperature.

[0016] Radiation heat exchange boundary: ; wherein, is the surface-to-surface radiation angle coefficient, considering the radiation exchange with the surrounding environment (such as the sky, adjacent buildings).

[0017] Thermal gradient calculation: ; Heat flux density calculation: ; wherein, is the fluid velocity, is the surrounding surface temperature; Temperature fluctuation frequency analysis: ; wherein, is the characteristic thickness, used to estimate the frequency of the material's response to temperature fluctuations.

[0018] Further, the process of verifying and optimizing the thermal response model includes: ; wherein, represents the parameters to be optimized (such as the internal surface convective heat transfer coefficient , surface absorption rate , external surface convective heat transfer coefficient ), by adjusting these parameters, the simulation results match the measured data, represents the i-th simulated temperature value calculated by the thermal response model based on the current parameters , the i-th measured temperature data point, n represents the total number of data points for calibration, including time series or spatial position combination, through Bayesian optimization adjustment , the error is minimized, and the process stops when the error converges or reaches the preset accuracy.

[0019] Further, the process of automatically grouping and panel generation of the three-dimensional geometric model includes: Traverse the point array of the three-dimensional geometric model, group according to the minimum grid vertex, traverse each array point in the generated point array, and group according to the rule of the minimum grid vertex, that is, adjacent vertices that constitute a minimum grid unit are divided into a group, which facilitates subsequent independent processing of each small area, and also conforms to the logic of curtain wall panel arrangement, because each group usually corresponds to the installation area of a panel, obtain a plurality of point array groups, extract the geometric features of each point array group, and adaptively match the panel according to the geometric features of each point array group, filter the panel from the preset panel library and embed it in the corresponding point array group position.

[0020] Further, the process of adaptively matching the panel according to the geometric features of each point array group includes: Obtain the geometric features of the point array group, obtain the constraint conditions of the panel, the constraint conditions include panel size limit, joint width requirement and structure support position, filter the panel that meets the constraint conditions from the preset panel library, perform similarity matching between the panel and the point array group, filter the panel with the highest similarity and embed it in the corresponding point array group position, and the calculation formula of the similarity matching is: ; wherein, simer represents the similarity, mdr represents the overlapping area of the panel and the point array group, mdq represents the area of the point array group, and mds represents the area of the panel.​

[0021] Further, the three-dimensional geometric model, the thermal characteristic data and the material characteristic data are fused to generate a three-dimensional curtain wall refined model, and a process of performing multi-physical field coupling verification on the three-dimensional curtain wall refined model includes: Based on the thermal response model, thermal characteristic data of each panel in the three-dimensional geometric model is output, material characteristic data of each panel in the three-dimensional geometric model is extracted, the thermal characteristic data and the material characteristic data of each panel are mapped to corresponding regions in the three-dimensional geometric model, and a three-dimensional curtain wall refined model is generated. Historical multi-source heterogeneous data of a target building is collected, a curtain wall heat transfer prediction model is constructed based on the historical multi-source heterogeneous data, multi-physical field coupling verification is performed on the three-dimensional curtain wall refined model according to the curtain wall heat transfer prediction model, model parameters of the three-dimensional curtain wall refined model are corrected according to a verification result, and a three-dimensional curtain wall refined model that has completed multi-physical field coupling verification is output. The three-dimensional curtain wall refined model is used to generate temperature field distribution data (such as a metal frame thermal bridge area temperature cloud map, a glass interlayer temperature gradient, and a per-hour surface temperature curve), heat flux density data (such as a unit area heat transfer rate and a glass-air interlayer heat flux density), stress and strain data (such as a thermal expansion stress caused by a temperature gradient, a maximum principal stress of a glass slab edge, a mechanical load response, and a structure deformation under wind pressure), and the like. The process of performing multi-physical field coupling verification on the three-dimensional curtain wall refined model according to the curtain wall heat transfer prediction model includes: Coupled physical fields, including heat conduction, convection, and radiation, are determined, thermal characteristic data and boundary condition inputs are extracted from the multi-source heterogeneous data and input into the three-dimensional curtain wall refined model and the curtain wall heat transfer prediction model, multi-physical field coupling simulation results are obtained according to the curtain wall heat transfer prediction model, and parameters corresponding to to-be-verified data, such as curtain wall surface temperature and heat flux density, are extracted from the multi-physical field coupling simulation results according to the three-dimensional curtain wall refined model output to-be-verified data; The parameters in the multi-physical field coupling simulation results are compared with the corresponding parameters in the to-be-verified data, root mean square errors of the parameters are obtained, a preset error threshold is set, parameters with root mean square errors greater than the error threshold are marked as to-be-corrected parameters, and for example, if a curtain wall surface temperature of a certain panel output by the three-dimensional curtain wall refined model is generally higher than a measured value output by the curtain wall heat transfer prediction model, the thermal conductivity of the curtain wall material may need to be reduced or the grid density of the panel region may need to be changed. Subsequently, sensitivity analysis is performed on the three-dimensional curtain wall refined model to determine the weight coefficients of each model parameter on the to-be-corrected parameter. Common model parameters include the thermal conductivity, specific heat capacity, emissivity, convective heat transfer coefficient, and grid density of the curtain wall material. Based on the weight coefficients of each model parameter on the to-be-corrected parameter, a multi-objective genetic algorithm is used to search for the optimal model parameter combination, which minimizes the root mean square error of the to-be-corrected parameter. The model parameters in the optimal model parameter combination are substituted into the three-dimensional curtain wall refined model, and multi-physics field coupling verification is performed again. The above process is repeated until the root mean square error meets the preset accuracy requirement, and the three-dimensional curtain wall refined model that completes the multi-physics field coupling verification is output.

[0022] Further, the process of constructing a curtain wall heat transfer prediction model based on historical multi-source heterogeneous data includes: A physical model is constructed through multi-physics field coupling analysis, including a heat transfer model (based on Fourier's law, establishing a heat conduction, convection, and radiation model of the curtain wall), a fluid flow model (using computational fluid dynamics (CFD) to simulate air flow and heat exchange on the surface of the curtain wall (such as natural convection, wind pressure driven air flow), and a structural stress model (analyzing material expansion or contraction stress caused by temperature changes). The coupling algorithm includes strong coupling: solving each physical field equation iteratively (such as heat transfer affecting material properties, material deformation reacting on the heat conduction path), weak coupling: solving each physical field in sequence (such as calculating the heat distribution first, then importing the temperature load into the structure analysis), and constructing a curtain wall heat transfer prediction model based on deep learning. Data preprocessing and feature extraction are performed on historical multi-source heterogeneous data to obtain thermal feature data and boundary conditions (such as thermal boundary: outdoor meteorological data (temperature, solar radiation), indoor air conditioning load, structural boundary: thermal expansion coefficient of curtain wall material, mechanical stress constraint). The thermal feature data and boundary conditions are used as the training set of the curtain wall heat transfer prediction model. The training set is input into the curtain wall heat transfer prediction model for training until the loss function training is stable, and the model parameters are saved. The thermal feature data and boundary conditions are input into the physical model, and the multi-physics field simulation results are output according to the physical model. The curtain wall heat transfer prediction model is tested through the multi-physics field simulation results until it meets the preset requirements, and the curtain wall heat transfer prediction model is output.

[0023] As shown in Figure 2 A curtain wall refined model generation system includes a cloud, which is communicatively connected with a data acquisition module, a curtain wall generation module, an optimization processing module, and a model generation module. The data acquisition module is used for collecting multi-source heterogeneous data of the target building and user intervention instructions, and performing data preprocessing on the multi-source heterogeneous data, wherein the multi-source heterogeneous data includes high-precision point cloud data, building line contour, environment data, material characteristic data, thermal response data, wind load data and temperature stress data; The curtain wall generation module is used for generating a curtain wall marginal line according to the building contour line, performing bidirectional dynamic adjustment of segmentation parameters on the curtain wall marginal line, and performing adaptive piecewise spline interpolation to generate a smooth curve, and generating a basic curtain wall base surface based on the smooth curve; The optimization processing module is used for constructing a basic three-dimensional geometric model according to the basic curtain wall base surface, generating an initial point array surface in combination with the segmentation points of the curtain wall marginal line, constructing a thermal response model in combination with the basic three-dimensional geometric model, the material characteristic data and the thermal response data, obtaining thermal characteristic data of each region, performing point array surface optimization of the basic three-dimensional geometric model according to the thermal characteristic data, generating a three-dimensional geometric model, and performing automatic grouping and panel generation operations on the three-dimensional geometric model; The model generation module is used for performing multi-modal feature fusion on the three-dimensional geometric model, the thermal characteristic data and the material characteristic data to generate a three-dimensional curtain wall refined model, and performing multi-physical field coupling verification on the three-dimensional curtain wall refined model.

[0024] Compared with the prior art, the present application has the following advantages: 1. Multi-source data fusion advantage: The present application widely collects multi-source heterogeneous data including high-precision point cloud data, building line contour, environment data, material characteristic data, thermal response data, wind load data and temperature stress data. Compared with the traditional method which only relies on a single or a small amount of data types, this comprehensive data fusion can more accurately reflect the various characteristics and behaviors of the curtain wall in the actual environment. For example, by integrating thermal response data and wind load data, the influence of wind on the heat transfer of the curtain wall can be comprehensively analyzed, so that the performance of the curtain wall under different climate conditions can be more accurately evaluated, and the model's simulation capability for complex real-world scenarios can be improved. 2. Structural and aesthetic advantage of dynamic segmentation and interpolation: The segmentation parameter of the curtain wall marginal line is dynamically adjusted in two directions, and the segmentation points are determined according to the wind load data, the temperature stress data and the user intervention instructions. The segmentation points can be encrypted in the structural stress concentration area to enhance the safety of the curtain wall structure and effectively reduce the risk of cracking caused by thermal bridge effect and wind force. At the same time, the adaptive piecewise spline interpolation generates a smooth curve, and the cubic spline or B-spline interpolation method is intelligently selected in the area with large curvature change to ensure that the generated basic curtain wall base surface has excellent smoothness, greatly meets the high requirements of architectural aesthetics on the curtain wall surface, and is beneficial to subsequent construction and processing, reducing material waste and installation difficulties caused by uneven curved surface. 3. Optimization advantage driven by thermal characteristics: A thermal response model is constructed by combining the basic three-dimensional geometric model, material characteristic data, and thermal response data. The point array surface of the basic three-dimensional geometric model is optimized according to the thermal characteristic data. In the high thermal gradient, high radiation, and other thermal sensitive areas, the grid density is analyzed based on the thermal characteristics, and the grid density is specifically improved. The heat flux density calculation error can be reduced from the conventional 8% to about 2%, which significantly improves the simulation accuracy of heat transfer. This helps to accurately identify the thermal weak link of the curtain wall and provides strong support for energy-saving design. For example, reasonable adjustment of material configuration in high thermal sensitive areas can reduce summer air conditioning load by 15%-20%, achieving effective reduction of building energy consumption. 4. Efficiency and cost advantage of automatic grouping and panel generation: The three-dimensional geometric model is subjected to automatic grouping and panel generation operation. The point array points are grouped by the minimum grid vertex, and the panel is adaptively matched from the preset panel library according to the point array group geometric characteristics. Compared with the traditional manual design and panel selection method, the design efficiency is greatly improved. At the same time, through the preset panel library, the demand for customized processing is reduced, the material utilization rate is improved, and the cost is effectively saved. 5. Multi-physical field coupling verification and model correction advantage: The multi-physical field coupling verification is performed on the three-dimensional curtain wall refined model, and the curtain wall heat transfer prediction model is constructed based on the historical multi-source heterogeneous data to verify and correct the refined model. Compared with the traditional single physical field analysis or the model lacking effective verification and correction mechanism, this method can more comprehensively consider the interaction between heat, structure, fluid and other multi-physical fields, identify potential problems such as sealant failure caused by temperature stress that cannot be found by traditional methods, and greatly improve the reliability and applicability of the model, providing more accurate basis for the maintenance and management of the whole life cycle of the curtain wall. 6. Prediction advantage of hybrid model construction: A physical model is constructed through multi-physical field coupling analysis, and a curtain wall heat transfer prediction model is constructed by combining deep learning. This hybrid model construction method combines the explainability of the physical model and the powerful learning ability and fast prediction ability of the deep learning model. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 A principle diagram of a curtain wall refined model generation method according to an embodiment of the present application.

[0026] Figure 2 A principle diagram of a curtain wall refined model generation system according to an embodiment of the present application. DETAILED DESCRIPTION

[0027] With reference to the drawings and in order to make the technical solutions in the embodiments of the present application clear, complete and easy to understand, obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the scope of the present application.

[0028] As shown in the drawings, Figure 1 A curtain wall refinement model generation method includes the following steps: Step s1: Collecting multi-source heterogeneous data of the target building and user intervention instructions, and performing data preprocessing on the multi-source heterogeneous data, the multi-source heterogeneous data including high-precision point cloud data, building line contour, environmental data, material characteristic data, thermal response data, wind load data and temperature stress data; Step s2: Generating a curtain wall marginal line according to the building contour line, performing bidirectional dynamic adjustment of the curtain wall marginal line segmentation parameters and adaptive piecewise spline interpolation, generating a smooth curve, and generating a basic curtain wall base surface based on the smooth curve; Step s3: Constructing a basic three-dimensional geometric model according to the basic curtain wall base surface, generating an initial point array surface in combination with the segmentation points of the curtain wall marginal line, constructing a thermal response model in combination with the basic three-dimensional geometric model, the material characteristic data and the thermal response data, obtaining thermal characteristic data of each region, optimizing the point array surface of the basic three-dimensional geometric model according to the thermal characteristic data, generating a three-dimensional geometric model, and performing automatic grouping and panel generation operations on the three-dimensional geometric model; Step s4: Performing multi-modal feature fusion on the three-dimensional geometric model, the thermal characteristic data and the material characteristic data, generating a three-dimensional curtain wall refinement model, and performing multi-physical field coupling verification on the three-dimensional curtain wall refinement model.

[0029] The process of collecting multi-source heterogeneous data of the target building includes: obtaining the building contour line, wind load sensitivity coefficient, temperature stress sensitivity coefficient, wind vibration coefficient, wind pressure height variation coefficient, building shape coefficient, material elastic modulus and material linear expansion coefficient of the target building according to the architectural design drawings, obtaining high-precision point cloud data of the target building by using a laser scanner or a drone, and synchronously collecting material reflection, refraction characteristics and thermal response data of the target building by using multi-spectral and thermal imaging technology. The process of data preprocessing on the collected multi-source heterogeneous data includes denoising processing to remove abnormal points caused by environmental interference and the like, and then calibration operation to ensure the consistency of data collected by different devices in spatial coordinates.

[0030] It needs to be further explained that in the specific implementation process, the curtain wall marginal line is generated according to the building contour line, the segmentation parameter of the curtain wall marginal line is dynamically adjusted in two directions, and the adaptive piecewise spline interpolation is carried out to generate a smooth curve, and the process of generating the basic curtain wall base surface based on the smooth curve includes: According to the building contour line, the target modeling element is selected by using the parametric design method to generate the curtain wall marginal line. Parametric design is a powerful design method that allows designers to control the geometry of the model by adjusting parameters. When generating the curtain wall marginal line, spline curves, geometric figures, etc. can be selected as target modeling elements. For example, when designing a curtain wall with a smooth curve shape, by setting control points and adjusting parameters, spline curves can easily follow the unique contour of the building, ensuring that the curtain wall marginal line perfectly matches the appearance of the building. For some regular-shaped curtain wall parts, such as rectangles, triangles, etc. geometric figures are more suitable. They can quickly build the basic framework of the curtain wall marginal line by simply defining parameters such as side length, angle, etc. According to the pre-processed wind load data and temperature stress data, as well as user intervention instructions, the segmentation parameter of the curtain wall marginal line is dynamically adjusted in two directions, and a number of segmentation points are marked on the curtain wall marginal line according to the completed two-way dynamic adjustment of the segmentation parameter. Key control points are selected on the curtain wall marginal line, and adaptive piecewise spline interpolation is carried out on the curtain wall marginal line according to the key control points to generate a smooth curve. The lofting operation is performed on a number of smooth curves, i.e. connecting these curves along a certain path to generate a continuous basic curtain wall base surface.

[0031] It needs to be further explained that in the specific implementation process, the process of marking a number of segmentation points on the curtain wall marginal line according to the segmentation parameter includes: The dynamic segmentation interval corresponding to each initial node is obtained, assuming that the dynamic segmentation interval corresponding to node Pi is di. This interval is relative to the reference segmentation interval and is the result of adjustment according to the specific conditions at each node, and it reflects the actual required segmentation density at the node. Starting from the initial node, one end of the curtain wall marginal line is selected as the starting point, the initial node closest to the end point is found as the first node to be processed, and the initial node is assumed to be P1. Taking P1 as the starting point, a new point Q1 is marked on the curtain wall marginal line along the direction of the marginal line by a distance d1 according to the dynamic segmentation interval d1 of P1. This point is the first segmentation point determined according to the dynamic segmentation interval of P1. Subsequently, other segmentation points are marked in turn. For the next initial node P2, the already marked point Q1 is taken as the starting point, and a new segmentation point Q2 is marked along the direction of the marginal line by a distance d2. In this way, each initial node Pi is processed in turn, with the previously marked segmentation point Qi-1 taken as the starting point, and the corresponding segmentation point Qi is marked along the direction of the marginal line by a distance di. When the boundary or end point of the curtain wall marginal line is encountered during the marking process, the marking method is adjusted. If the remaining length of the marginal line is less than the dynamic segmentation interval di of the current node, the remaining length is taken as the segmentation interval, and the last segmentation point is marked so that it is located at the end point of the marginal line. If it is found during the marking process that the segmentation points corresponding to two adjacent initial nodes may coincide or have too small an interval (less than a preset minimum interval threshold), the positions of the segmentation points are fine-tuned, and the segmentation interval of one of the nodes is appropriately increased to ensure that the segmentation points have a reasonable interval and accurately reflect the characteristics and design requirements of the curtain wall marginal line.

[0032] It should be further noted that, in the specific implementation process, the process of bidirectional dynamic adjustment of the segmentation parameters of the curtain wall marginal line according to the preprocessed wind load data and temperature stress data and user intervention instructions includes: presetting a reference segmentation interval, a structure response weight, and a curvature weight, and generating an initial point set along the curtain wall marginal line at a uniform interval according to the reference segmentation interval performing differential operation on each initial node in the initial point set to obtain the curvature value k of each initial node, mapping the wind load data and the temperature stress data to the initial point set for multi-field coupling analysis to obtain the total stress coefficient of each initial node, and obtaining the dynamic segmentation interval corresponding to each initial node according to the total stress coefficient and the curvature value of each initial node and the structure response weight and the curvature weight; The calculation formula for obtaining the dynamic segmentation interval of each initial node is: ; ; ; ; ; wherein, represents a dynamic segmentation interval of an initial node i, represents a reference segmentation interval, represents a structural response weight, represents a curvature weight, represents a total stress coefficient of the initial node i, represents a curvature value of the initial node i, represents a wind load stress coefficient, represents a temperature stress coefficient, represents a wind load sensitivity coefficient, represents a temperature stress sensitivity coefficient, represents a wind vibration coefficient, represents a wind pressure height variation coefficient, represents a building shape coefficient, represents a material elastic modulus, represents a material linear expansion coefficient, represents a temperature difference, represents a local annual temperature extreme value; According to the user intervention instruction, the segmentation interval fixed threshold interval corresponding to each initial node is obtained, it is judged whether the dynamic segmentation interval corresponding to each initial node is located in the corresponding segmentation interval fixed threshold interval, if yes, the dynamic segmentation interval of the initial node is marked as the best segmentation interval, if not, the best segmentation interval of the initial node is set according to the segmentation interval fixed threshold interval, the maximum segmentation interval in the segmentation interval fixed threshold interval is selected as the best segmentation interval, the segmentation parameters are constructed according to the best segmentation interval corresponding to each initial node, the density of the segmentation point is dynamically adjusted based on the curvature analysis, the accuracy of the surface fitting is ensured, then the influence of structural factors such as wind load and temperature change on the curtain wall is further considered, so as to optimize the position of the segmentation point, reduce stress concentration, improve the stability of the curtain wall fine model structure, and improve the accuracy of subsequent analysis.

[0033] It needs to be further explained that, in the specific implementation process, the process of selecting the key control point on the curtain wall marginal line and adaptively segmenting the spline interpolation of the curtain wall marginal line to generate a smooth curve includes: The curvature value and the total stress coefficient of each segmentation point on the curtain wall marginal line are weighted and averaged to obtain the key coefficient of each segmentation point, a key coefficient threshold is preset, the key coefficient of each segmentation point is compared with the key coefficient threshold, the segmentation point with the key coefficient greater than the key coefficient threshold is marked as a key control point, the curvature change analysis and the total stress change analysis are performed on the adjacent key control points, the curvature change coefficient and the total stress change coefficient between the adjacent key control points are obtained. Preset thresholds corresponding to the curvature variation coefficient and the total stress variation coefficient, and determine whether the curvature variation coefficient and the total stress variation coefficient between adjacent key control points are both less than the corresponding thresholds. If both are less than, then use the cubic spline interpolation method to generate a smooth curve between adjacent key control points. Cubic spline interpolation is a commonly used spline interpolation method. It uses a cubic polynomial to approximate the curve in each small interval and ensures the continuity of the first-order and second-order derivatives of the curve at the control points, thereby generating a smooth curve. If both are not less than, then use the B-spline interpolation method to generate a smooth curve between adjacent key control points. B-spline interpolation controls the local shape of the curve through node vectors, supports piecewise polynomials, and is suitable for complex geometry. Perform error analysis and iterative optimization on the smooth curve between adjacent key control points, and extract several discrete points on the curtain wall edge line between adjacent key control points , calculate the minimum distance to the smooth curve between adjacent key control points , set the maximum error threshold, if If the error is greater than the maximum error threshold, a node insertion algorithm (such as the Cox-deBoor algorithm) is used to add new nodes and control points and re-interpolate. The optimization is stopped when the error is less than or equal to the maximum error threshold or the number of iterations reaches the preset upper limit.

[0034] In areas with small changes in curvature and total stress, the relatively simple cubic spline interpolation method is used, avoiding the additional computational effort required by the more complex B-spline interpolation method. Cubic spline interpolation is fast and requires fewer computational resources, improving efficiency and reducing computational time and cost while ensuring curve quality. In areas with large changes in curvature and total stress, B-spline interpolation, while relatively complex, is used. However, it effectively handles complex curve shapes and avoids repeated adjustments and corrections that would be required if simpler methods could not accurately fit the curve. Overall, this can save time and resources and improve design and production efficiency.

[0035] Perform curvature change analysis and total stress change analysis on adjacent key control points, and obtain the calculation formula of curvature change coefficient and total stress change coefficient between adjacent key control points: Set adjacent critical control points and The total stress coefficients at and , the curvature values ​​are and ; ; ; wherein, represents a curvature variation coefficient, represents a total stress variation coefficient.

[0036] It should be further explained that, in the specific implementation process, in combination with the segmentation points of the curtain wall marginal line, an initial point array surface is generated, the point array surface optimization of the basic three-dimensional geometric model is performed, and the process of generating the three-dimensional geometric model includes: mapping the surface parameter space of the basic curtain wall base surface to the physical coordinate system to construct a basic three-dimensional geometric model, using a nearest point projection algorithm to map the segmentation points of the curtain wall marginal line to the basic curtain wall base surface to generate an initial point array surface, performing Delaunay triangulation on the initial point array surface to generate a plurality of initial triangular meshes; The thermal characteristic data includes thermal gradient, heat flux density, temperature fluctuation frequency, the thermal characteristic data of each region is taken as an evaluation index, the index weight of the evaluation index is set, and the membership matrix of each region for a preset thermal sensitivity level is obtained through fuzzy comprehensive evaluation; The thermal sensitivity level of each region is obtained according to the membership matrix and the index weight, a thermal sensitivity level threshold is preset, a region with a thermal sensitivity level greater than the thermal sensitivity level threshold is marked as a thermal sensitivity region, grid density analysis is performed on the thermal characteristic data of the thermal sensitivity region, the optimized grid density of the thermal sensitivity region is obtained, and a plurality of initial triangular meshes in the thermal sensitivity region in the basic three-dimensional geometric model are re-divided according to the optimized grid density to generate a three-dimensional geometric model.

[0037] It should be further explained that, in the specific implementation process, the process of obtaining the thermal sensitivity level of each region according to the membership matrix and the index weight includes: The index weight of the evaluation index and the membership matrix are fused through a formula to obtain a fuzzy comprehensive evaluation matrix of the evaluation index, the membership degrees of each region for different thermal sensitivity levels are obtained according to the fuzzy comprehensive evaluation matrix, the thermal sensitivity level with the highest membership degree corresponding to each region is screened out, and the thermal sensitivity level with the highest membership degree corresponding to each region is taken as the thermal sensitivity level of each region; wherein, the formula is: ; wherein, is the fuzzy comprehensive evaluation matrix of the evaluation index, is the index weight of the evaluation index, is the membership matrix, indicates that the elements at the corresponding positions of the weight matrix of the evaluation index and the membership matrix are multiplied, is a weighting parameter for controlling the balance between the weight matrix and the membership matrix in the fuzzy comprehensive evaluation matrix of the evaluation index.

[0038] The process of grid density analysis according to the thermal characteristic data of the heat-sensitive area to obtain the optimized grid density of the heat-sensitive area includes: ; Wherein, is the optimized grid size, represents the initial grid size, represents the thermal gradient, represents the maximum thermal gradient of the region, represents the heat flux density, represents the thermal conductivity of the material, represents the temperature fluctuation frequency (unit: Hz, such as the fluctuation caused by periodic solar radiation), represents the frequency correlation coefficient (typical value 0.1~0.3, which needs to match the time step), 、 and represent the weight factor, the above formulas are dimensionless values, the formula is obtained by software simulation of a large amount of data to obtain a formula closest to the real situation, and the preset parameters and preset thresholds in the formula are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation; For steady-state problems, the formula is mainly and For transient problems, the formula also considers At the same time, it is compatible with the needs of steady-state and transient thermal analysis, realizes the precise matching of grid density and thermal characteristics, and significantly improves the reliability of the prediction of curtain wall heat transfer.

[0039] It needs to be further explained that in the specific implementation process, the process of constructing a thermal response model combining the basic three-dimensional geometric model, material characteristic data and thermal response data to obtain the thermal characteristic data of each area includes: Collecting the material characteristic data and thermal response data of each initial triangular grid in the basic three-dimensional geometric model, performing grid optimization preprocessing on each initial triangular grid according to the material characteristic data of each initial triangular grid, obtaining each triangular grid after grid optimization preprocessing, constructing a thermal response model combining the material characteristic data and thermal response data of each triangular grid, verifying and optimizing the thermal response model, and obtaining the thermal response model after verification and optimization; According to the thermal response model, output the thermal characteristic data of each area of the basic three-dimensional geometric model.

[0040] Wherein, the process of performing grid optimization preprocessing on each initial triangular grid according to the material characteristic data of each initial triangular grid includes: Adjusting the grid size of each initial triangular grid according to the material characteristic data of each initial triangular grid, and the calculation formula is: ; wherein, represents the adjusted grid size, represents the unadjusted grid size; It should be further noted that the mathematical model system required for constructing the thermal response model in combination with the material property data and thermal response data of each triangular mesh is as follows: Transient heat conduction control equation: ; wherein, is the temperature field, t is time, is the density, is the specific heat capacity, is the thermal conductivity, is the heat source (such as solar radiation heat production).

[0041] External surface solar radiation boundary: ; wherein, is the surface absorptivity, is the solar irradiance, is the emissivity, is the Stefan-Boltzmann constant, is the external surface convective heat transfer coefficient, is the ambient temperature.

[0042] Internal surface convective boundary: ; wherein, is the internal surface convective heat transfer coefficient, is the indoor air temperature.

[0043] Radiation heat exchange boundary: ; wherein, is the surface-to-surface radiation angle coefficient, considering the radiation exchange with the surrounding environment (such as the sky, adjacent buildings).

[0044] Thermal gradient calculation: ; Heat flux density calculation: ; wherein, is the fluid velocity, is the surrounding surface temperature; Temperature fluctuation frequency analysis: ; wherein, is the characteristic thickness, used to estimate the frequency of the material's response to temperature fluctuations.

[0045] It should be further explained that the process of verifying and optimizing the thermal response model includes: ; wherein, represents the parameters to be optimized (such as the internal surface convective heat transfer coefficient , surface absorption rate , external surface convective heat transfer coefficient ), by adjusting these parameters, the simulation results match the measured data, represents the i-th simulated temperature value calculated by the thermal response model based on the current parameters , the i-th measured temperature data point, n represents the total number of data points for calibration, including time series or spatial position combinations, and the parameters are continuously optimized and adjusted by Bayes , so as to minimize the error, and the process stops when the error converges or reaches the preset accuracy.

[0046] It should be further explained that in the specific implementation process, the process of automatically grouping and panel generating for the three-dimensional geometric model includes: Traverse the point array of the three-dimensional geometric model, group according to the minimum grid vertex, traverse each array point in the generated point array, and group according to the rule of the minimum grid vertex, that is, divide the adjacent vertices constituting the minimum grid unit into a group, which facilitates subsequent independent processing of each small area, and also conforms to the logic of curtain wall panel arrangement, because each group usually corresponds to the installation area of a panel, obtains a plurality of point array groups, extracts the geometric features of each point array group, and adaptively matches the panel according to the geometric features of each point array group, and filters the panel from the preset panel library and embeds the corresponding point array group position.

[0047] It should be further explained that in the specific implementation process, the process of adaptively matching the panel according to the geometric features of each point array group includes: Obtain the geometric features of the point array group, obtain the constraint conditions of the panel, the constraint conditions include panel size limit, joint width requirement and structural support position, filter the panel that meets the constraint conditions from the preset panel library, perform similarity matching between the panel and the point array group, filter the panel with the highest similarity and embed it in the corresponding point array group position, and the calculation formula of the similarity matching is: ; wherein, simer represents the similarity, mdr represents the overlapping area of the panel and the point array group, mdq represents the area of the point array group, and mds represents the area of the panel.​

[0048] It needs to be further explained that, in the specific implementation process, the three-dimensional geometric model, the thermal characteristic data and the material characteristic data are fused to generate a three-dimensional curtain wall refined model, and the process of verifying the three-dimensional curtain wall refined model by multi-physical field coupling includes: Based on the thermal response model, the thermal characteristic data of each panel in the three-dimensional geometric model is output, the material characteristic data of each panel in the three-dimensional geometric model is extracted, the thermal characteristic data and the material characteristic data of each panel are mapped to the corresponding area in the three-dimensional geometric model, and a three-dimensional curtain wall refined model is generated; The historical multi-source heterogeneous data of the target building is collected, a curtain wall heat transfer prediction model is constructed based on the historical multi-source heterogeneous data, the three-dimensional curtain wall refined model is verified by multi-physical field coupling according to the curtain wall heat transfer prediction model, the model parameters of the three-dimensional curtain wall refined model are corrected according to the verification result, and a three-dimensional curtain wall refined model that has completed the multi-physical field coupling verification is output. The three-dimensional curtain wall refined model is used to generate temperature field distribution data (such as metal frame thermal bridge area temperature cloud map, glass interlayer temperature gradient, hourly surface temperature curve), heat flux density data (such as unit area heat transfer rate, glass-air interlayer heat flux density), stress and strain data (such as thermal expansion stress caused by temperature gradient, maximum principal stress of glass slab edge, mechanical load response, structural deformation under wind pressure) and the like; Among them, the process of verifying the three-dimensional curtain wall refined model by multi-physical field coupling according to the curtain wall heat transfer prediction model includes: Determine the coupled physical fields, including heat conduction, convection, radiation, etc., extract the thermal characteristic data and boundary condition from the multi-source heterogeneous data and input them into the three-dimensional curtain wall refined model and the curtain wall heat transfer prediction model, obtain the multi-physical field coupling simulation results according to the curtain wall heat transfer prediction model, output the to-be-verified data according to the three-dimensional curtain wall refined model, and extract the parameters corresponding to the to-be-verified data from the multi-physical field coupling simulation results, such as curtain wall surface temperature, heat flux density, etc.; Compare the parameters in the multi-physical field coupling simulation results with the corresponding parameters in the to-be-verified data, obtain the root mean square error corresponding to the parameters, preset an error threshold, mark the parameters with a root mean square error greater than the error threshold as to-be-corrected parameters, for example, if the curtain wall surface temperature of a certain panel output by the three-dimensional curtain wall refined model is generally higher than the measured value output by the curtain wall heat transfer prediction model, the thermal conductivity of the curtain wall material may need to be reduced or the grid density of the panel belonging to the area may need to be changed; Subsequently, sensitivity analysis is performed on the three-dimensional curtain wall refined model to determine the weight coefficients of each model parameter on the to-be-corrected parameter. Common model parameters include the thermal conductivity of curtain wall materials, specific heat capacity, emissivity, convective heat transfer coefficient, grid density, etc. Based on the weight coefficients of each model parameter on the to-be-corrected parameter, a multi-objective genetic algorithm is used to search for the optimal model parameter combination, which minimizes the root mean square error of the to-be-corrected parameter. The model parameters in the optimal model parameter combination are substituted into the three-dimensional curtain wall refined model, and multi-physics field coupling verification is performed again. The above process is repeated until the root mean square error meets the preset accuracy requirement, and the three-dimensional curtain wall refined model that completes the multi-physics field coupling verification is output.

[0049] The process of constructing the curtain wall heat transfer prediction model based on historical multi-source heterogeneous data includes: A physical model is constructed through multi-physics field coupling analysis, including a heat transfer model (based on Fourier's law, establishing a heat conduction, convection, and radiation model for the curtain wall), a fluid flow model (using computational fluid dynamics (CFD) to simulate air flow and heat exchange on the curtain wall surface (such as natural convection, wind pressure driven air flow), and a structural stress model (analyzing material expansion or contraction stress caused by temperature changes). The coupling algorithm includes strong coupling: solving each physical field equation iteratively (such as heat transfer affecting material properties, material deformation reacting on the heat conduction path), weak coupling: solving each physical field in sequence (such as calculating the heat distribution first, then importing the temperature load into the structure analysis), and constructing a curtain wall heat transfer prediction model based on deep learning. Data preprocessing and feature extraction are performed on historical multi-source heterogeneous data to obtain thermal feature data and boundary conditions (such as thermal boundary: outdoor meteorological data (temperature, solar radiation), indoor air conditioning load, structural boundary: thermal expansion coefficient of curtain wall materials, mechanical stress constraint). The thermal feature data and boundary conditions are used as the training set of the curtain wall heat transfer prediction model. The training set is input into the curtain wall heat transfer prediction model for training until the loss function training is stable, and the model parameters are saved. The thermal feature data and boundary conditions are input into the physical model, and the multi-physics field simulation results are output according to the physical model. The curtain wall heat transfer prediction model is tested through the multi-physics field simulation results until it meets the preset requirements, and the curtain wall heat transfer prediction model is output.

[0050] Constructing a curtain wall heat transfer prediction model is a complex process that involves multiple steps such as model selection, training, verification, and testing. The following is a detailed supplementary explanation of this process: A convolutional neural network (CNN), suitable for time series analysis, was selected as the deep learning architecture, and the cross-entropy loss function was chosen as the optimization objective. The prepared training set was then fed into the selected deep learning model to begin training. During training, the weights were continuously updated using the backpropagation algorithm, gradually reducing the loss function until a steady state was reached. During this process, techniques such as early stopping were utilized to prevent overfitting. In addition to the basic training process, various model parameters, including the learning rate, batch size, and regularization coefficient, were tuned using grid search.

[0051] After model training is complete and parameter adjustments are complete, a final evaluation is performed on the test set to obtain the model's evaluation results. These results include classification metrics such as accuracy, recall, and F1 score. Based on the evaluation results on the test set, it is determined whether the model meets the expected standards. If the requirements are met, the model parameters are saved and prepared for deployment. If not, it is necessary to return to a previous stage to re-examine issues such as data quality, model structure, or training strategy.

[0052] like Figure 2 As shown, a curtain wall refined model generation system includes a cloud, wherein the cloud is communicatively connected to a data acquisition module, a curtain wall generation module, an optimization processing module, and a model generation module; The data acquisition module is used to collect multi-source heterogeneous data of the target building and user intervention instructions, and perform data preprocessing on the multi-source heterogeneous data. The multi-source heterogeneous data includes high-precision point cloud data, building line outlines, environmental data, material property data, thermal response data, wind load data, and temperature stress data; The curtain wall generation module is used to generate curtain wall edge lines according to the building outline, dynamically adjust the segmentation parameters of the curtain wall edge lines in both directions, and perform adaptive segmented spline interpolation to generate smooth curves. Based on the smooth curves, the basic curtain wall surface is generated. The optimization processing module is used to construct a basic 3D geometric model based on the basic curtain wall surface, generate an initial lattice surface based on the segmentation points of the curtain wall edge line, construct a thermal response model based on the basic 3D geometric model, material property data, and thermal response data, obtain thermal characteristic data of each area, optimize the lattice surface of the basic 3D geometric model based on the thermal characteristic data, generate a 3D geometric model, and perform automated grouping and panel generation operations on the 3D geometric model; The model generation module is used to perform multimodal feature fusion of the three-dimensional geometric model, thermal characteristic data and material property data to generate a three-dimensional curtain wall refined model and perform multi-physical field coupling verification on the three-dimensional curtain wall refined model.

[0053] The above examples are only used to illustrate the technical method of the present application but not limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present application.

Claims

1. A curtain wall refined model generation method, characterized in that: The following steps are involved: Step s1: Collect multi-source heterogeneous data of the target building and user intervention instructions, and perform data preprocessing on the multi-source heterogeneous data, wherein the multi-source heterogeneous data includes high-precision point cloud data, building outlines, environmental data, material property data, thermal response data, wind load data, and temperature stress data; Step s2: Generate curtain wall edge lines according to the building outline, dynamically adjust segmentation parameters of the curtain wall edge lines in both directions and perform adaptive segmented spline interpolation to generate smooth curves, and generate basic curtain wall surfaces based on the smooth curves; Step s3: Construct a basic 3D geometric model based on the basic curtain wall surface. Combine the segmentation points of the curtain wall edge line to generate an initial lattice surface. Combine the basic 3D geometric model, material property data, and thermal response data to construct a thermal response model. Obtain thermal characteristic data of each area. Optimize the lattice surface of the basic 3D geometric model based on the thermal characteristic data to generate a 3D geometric model. Automatically group and panelize the 3D geometric model. Step s4: Perform multimodal feature fusion on the three-dimensional geometric model, thermal feature data, and material property data to generate a three-dimensional curtain wall refined model, and perform multi-physics field coupling verification on the three-dimensional curtain wall refined model.

2. The method for generating a curtain wall refined model according to claim 1, characterized in that: Generate curtain wall edge lines, dynamically adjust segmentation parameters of curtain wall edge lines in both directions, and perform adaptive segmented spline interpolation to generate smooth curves. The process of generating basic curtain wall surfaces based on smooth curves includes: Based on the building outline, a parametric design method is used to select target modeling elements to generate curtain wall edge lines. The segmentation parameters of the curtain wall edge lines are dynamically adjusted in both directions based on pre-processed wind load data, temperature stress data, and user intervention instructions. Several segmentation points are marked on the curtain wall edge lines according to the segmentation parameters. Key control points are selected from the several segmentation points. Based on the key control points, the curtain wall edge lines are adaptively interpolated with segmented splines to generate smooth curves. Perform lofting operations on several smooth curves to generate the basic surface of the basic curtain wall.

3. The method for generating a curtain wall refined model according to claim 2, characterized in that: The process of bidirectionally dynamically adjusting the segmentation parameters of the curtain wall edge line based on pre-processed wind load data, temperature stress data, and user intervention instructions includes: The benchmark segmentation spacing, structural response weight, and curvature weight are preset. An initial point set is uniformly generated along the curtain wall edge line based on the benchmark segmentation spacing. The curvature value of each initial node is obtained. Wind load data and temperature stress data are mapped to the initial point set for multi-field coupling analysis. The total stress coefficient of each initial node is obtained. Based on the total stress coefficient and curvature value of each initial node and the structural response weight and curvature weight, the dynamic segmentation spacing corresponding to each initial node is obtained. According to the user intervention instruction, the fixed threshold interval of the segmentation spacing corresponding to each initial node is obtained, and it is determined whether the dynamic segmentation spacing corresponding to each initial node is within the corresponding fixed threshold interval of the segmentation spacing. If it is, the dynamic segmentation spacing of the initial node is marked as the optimal segmentation spacing. If it is not, the optimal segmentation spacing of the initial node is set according to the fixed threshold interval of the segmentation spacing, and the segmentation parameters are constructed according to the optimal segmentation spacing corresponding to each initial node.

4. The method for generating a curtain wall refined model according to claim 3, characterized in that: The process of selecting key control points and performing adaptive segmented spline interpolation on the curtain wall edge line based on the key control points to generate a smooth curve includes: Perform weighted averaging on the curvature values ​​and total stress coefficients of each segmentation point on the curtain wall edge line to obtain the critical coefficient of each segmentation point. Preset a critical coefficient threshold, mark the segmentation points with a critical coefficient greater than the critical coefficient threshold as key control points, perform curvature change analysis and total stress change analysis on adjacent key control points, and obtain the curvature change coefficient and total stress change coefficient between adjacent key control points. Preset thresholds corresponding to the curvature variation coefficient and the total stress variation coefficient, and determine whether the curvature variation coefficient and the total stress variation coefficient between adjacent key control points are both less than the corresponding thresholds. If both are less than, use the cubic spline interpolation method to generate a smooth curve between the adjacent key control points. If both are not less than, use the B-spline interpolation method to generate a smooth curve between the adjacent key control points. Error analysis and iterative optimization are performed on the smooth curves between adjacent key control points.

5. The method for generating a curtain wall refined model according to claim 4, characterized in that: Combined with the segmentation points of the curtain wall edge line, an initial lattice surface is generated, and the lattice surface of the basic 3D geometric model is optimized. The process of generating a 3D geometric model includes: Mapping the segmentation points of the curtain wall edge line to the basic surface of the basic curtain wall to generate an initial lattice surface, and triangulating the initial lattice surface to generate a number of initial triangular meshes; Based on the thermal characteristic data of each area as the evaluation index, the indicator weight of the evaluation index is set, and the membership matrix of each area to the preset thermal sensitivity level is obtained through fuzzy comprehensive evaluation; The thermal sensitivity level of each area is obtained according to the membership matrix and indicator weights, and a thermal sensitivity level threshold is preset. The areas with thermal sensitivity levels greater than the thermal sensitivity level threshold are marked as thermally sensitive areas. The grid density of the thermally sensitive areas is analyzed based on thermal characteristics to obtain the optimized grid density of the thermally sensitive areas. According to the optimized grid density, several initial triangular meshes in the thermally sensitive areas of the basic three-dimensional geometric model are re-divided to generate a three-dimensional geometric model.

6. The method for generating a curtain wall refined model according to claim 5, characterized in that: The process of building a thermal response model by combining the basic 3D geometric model, material property data, and thermal response data to obtain thermal characteristic data for each area includes: Collecting material property data and thermal response data of each initial triangular mesh in the basic three-dimensional geometric model, performing mesh optimization preprocessing on each initial triangular mesh according to the material property data of each initial triangular mesh, obtaining each triangular mesh after mesh optimization preprocessing, constructing a thermal response model based on the material property data and thermal response data of each triangular mesh, verifying and optimizing the thermal response model, and obtaining a thermal response model that has completed verification and optimization; The thermal characteristic data of each area of ​​the basic three-dimensional geometric model are output according to the thermal response model.

7. The method for generating a curtain wall refined model according to claim 6, characterized in that: The process of automatically grouping and paneling 3D geometry models includes: The lattice points of the 3D geometric model are traversed and grouped according to the minimum mesh vertices to obtain several lattice groups. The geometric features of each lattice group are extracted, and adaptive panel matching is performed based on the geometric features of each lattice group. Panels are selected from the preset panel library and embedded into the corresponding lattice group positions.

8. The method for generating a curtain wall refined model according to claim 7, characterized in that: The process of generating a refined 3D curtain wall model and performing multi-physics coupling verification on the refined 3D curtain wall model includes: Outputting thermal characteristic data of each panel in the 3D geometric model based on the thermal response model, extracting material characteristic data of each panel in the 3D geometric model, mapping the thermal characteristic data and material characteristic data of each panel to the corresponding area in the 3D geometric model, and generating a 3D curtain wall refined model; Collect historical multi-source heterogeneous data of the target building, build a curtain wall heat transfer prediction model based on the historical multi-source heterogeneous data, perform multi-physics field coupling verification on the three-dimensional curtain wall refined model according to the curtain wall heat transfer prediction model, and modify the model parameters of the three-dimensional curtain wall refined model according to the verification results.

9. The method for generating a curtain wall refined model according to claim 8, characterized in that: The process of building a curtain wall heat transfer prediction model based on historical multi-source heterogeneous data includes: A physical model is constructed through multi-physics field coupling analysis, a curtain wall heat transfer prediction model is constructed based on deep learning, data preprocessing and feature extraction are performed on historical multi-source heterogeneous data, thermal feature data and boundary conditions are obtained, the thermal feature data and boundary conditions are used as a training set for the curtain wall heat transfer prediction model, the training set is input into the curtain wall heat transfer prediction model for training until the loss function training is stable, and the model parameters are saved. At the same time, the thermal feature data and boundary conditions are input into the physical model, and multi-physics field simulation results are output according to the physical model. The curtain wall heat transfer prediction model is tested by the multi-physics field simulation results until it meets the preset requirements, and the curtain wall heat transfer prediction model is output.

10. A curtain wall refined model generation system, specifically applied to a curtain wall refined model generation method according to any one of claims 1 to 9, characterized in that: It includes a cloud, wherein the cloud is communicatively connected to a data acquisition module, a curtain wall generation module, an optimization processing module and a model generation module; The data acquisition module is used to collect multi-source heterogeneous data of the target building and user intervention instructions, and perform data preprocessing on the multi-source heterogeneous data. The multi-source heterogeneous data includes high-precision point cloud data, building line outlines, environmental data, material property data, thermal response data, wind load data, and temperature stress data; The curtain wall generation module is used to generate curtain wall edge lines according to the building outline, dynamically adjust the segmentation parameters of the curtain wall edge lines in both directions, and perform adaptive segmented spline interpolation to generate smooth curves. Based on the smooth curves, the basic curtain wall surface is generated. The optimization processing module is used to construct a basic 3D geometric model based on the basic curtain wall surface, generate an initial lattice surface based on the segmentation points of the curtain wall edge line, construct a thermal response model based on the basic 3D geometric model, material property data, and thermal response data, obtain thermal characteristic data of each area, optimize the lattice surface of the basic 3D geometric model based on the thermal characteristic data, generate a 3D geometric model, and perform automated grouping and panel generation operations on the 3D geometric model; The model generation module is used to perform multimodal feature fusion of the three-dimensional geometric model, thermal characteristic data and material property data to generate a three-dimensional curtain wall refined model and perform multi-physical field coupling verification on the three-dimensional curtain wall refined model.

Images