A method, medium and system for predicting thermal insulation effect in civil engineering design stage

Abstract

The present invention provides a method, medium and system for predicting the thermal insulation effect in the civil engineering design stage, which belongs to the field of civil engineering thermal insulation technology, and includes: establishing a civil engineering three-dimensional finite element model based on civil engineering design drawing information, and generating a series of scale models of size ratios based on the model. The CUDA parallel computing capability of the GPU is used to calculate the thermal performance parameters of each component of the minimum scale model, including thermal resistance and heat transfer coefficient. Next, the temperature field of the minimum scale model is established, and the temperature field is sparsely processed using the parallel algorithm of the Thrust library. The Gaussian mixture model is used to cluster the sparse matrix, and the thermal bridge area and the non-thermal bridge area are divided, and the heat flux density of each component in the non-thermal bridge area and the local heat flux density in the thermal bridge area are calculated respectively. Representative components are selected and mapped to a 1.00 times scale model to obtain the thermal performance parameters of the model and the thermal insulation effect index in the civil engineering design stage.

Description

Technical Field

[0001] The present invention belongs to the technical field of civil engineering thermal insulation, and in particular, relates to a thermal insulation effect prediction method, medium and system in a civil engineering design stage. Background Art

[0002] In the field of building energy conservation, the quality of thermal insulation directly affects the energy consumption of buildings and the quality of indoor environment. Therefore, it is necessary to accurately evaluate and optimize the thermal insulation performance in the civil engineering design stage. At present, three-dimensional finite element analysis is the mainstream method for evaluating the thermal insulation effect of civil engineering. By establishing a three-dimensional finite element model containing information such as geometric dimensions and material parameters, the temperature field distribution inside the building can be simulated, and then the thermal performance indicators of each building component, such as heat flux density and thermal resistance, can be calculated. Based on these thermal performance indicators, the overall thermal insulation effect can be predicted. This thermal insulation effect evaluation method based on three-dimensional finite element analysis is widely used in various construction projects, including residential buildings, office buildings, hospitals, schools, etc. Taking residential buildings as an example, in the design stage, architects need to comprehensively consider factors such as the orientation, floor height, and exterior wall materials of the building, and select appropriate thermal insulation measures to maximize the thermal insulation effect while meeting economic and aesthetic requirements. Through finite element analysis, architects can quickly evaluate the thermal insulation performance of different design schemes and provide a basis for determining the final design scheme.

[0003] However, the existing insulation effect evaluation methods based on 3D finite element analysis have some problems and limitations. Since the 3D finite element model usually has a huge number of grids, if the entire model is directly refined, it will consume a lot of computing resources and be inefficient. Especially for large-scale construction projects, the amount of calculation often exceeds the capacity of ordinary computers. Summary of the invention

[0004] In view of this, the present invention provides a thermal insulation effect prediction method, medium and system in the civil engineering design stage, which can solve the technical problem that if the existing thermal insulation effect based on three-dimensional finite element analysis is directly calculated for the entire model, a large amount of computing resources will be consumed, resulting in low efficiency.

[0005] The present invention is achieved in that:

[0006] A first aspect of the present invention provides a method for predicting thermal insulation effect in a civil engineering design stage, comprising the following steps:

[0007] S10, establishing a civil engineering three-dimensional finite element model according to the structural information and material parameters of the civil engineering design drawings, and establishing multiple scale models with sequential size ratios based on the civil engineering three-dimensional finite element model, wherein the finite element mesh sizes of the multiple scale models are equal, and the sequential size ratio is a specified number of multiple size ratios from 0.01 times the size ratio to 1.00 times the size;

[0008] S20, using CUDA parallel computing capabilities to calculate thermal performance parameters of each building component of the scale model of the minimum size ratio on the GPU, including thermal resistance and heat transfer coefficient;

[0009] S30, continue to use the numerical simulation method on the GPU to establish the temperature field of the proportional model with the minimum size ratio;

[0010] S40, sparseizing the temperature field matrix to obtain a sparse matrix, and dividing the sparse matrix into thermal bridge areas and non-thermal bridge areas according to temperature distribution;

[0011] S50, transmitting the clustering result back to the CPU, identifying and extracting multiple thermal bridge areas existing in the 3D model;

[0012] S60, switching back to the GPU, and calculating the heat flux density of each building component in the non-thermal bridge area and the local heat flux density in the thermal bridge area according to the temperature field distribution inside and outside the building;

[0013] S70, clustering the thermal performance parameters of each building component of the multiple scale models to obtain multiple cluster centers, and obtaining multiple building components with the shortest distances to the cluster centers as representative building components;

[0014] S80, using representative building components, continuously mapping them to corresponding components of a more proportional scale model, and finally integrating them into a civil engineering three-dimensional finite element model of 1.00 times the size, to obtain the heat flux density of each building component in the non-thermal bridge area of ​​the civil engineering three-dimensional finite element model and the local heat flux density in the thermal bridge area;

[0015] S90. Finally, the heat flux density of each building component in the non-thermal bridge area of ​​the civil engineering three-dimensional finite element model and the local heat flux density in the thermal bridge area are transmitted back to the CPU, input into a pre-trained thermal insulation effect prediction model, and the thermal insulation effect index of the civil engineering design stage is obtained, and output to the designer.

[0016] Among them, the method of dividing the sparse matrix into thermal bridge areas and non-thermal bridge areas according to temperature distribution is the Gaussian mixture model clustering algorithm.

[0017] The formula for calculating heat flux density is: q=λ*ΔT / d, where q is heat flux density, λ is material thermal conductivity, ΔT is temperature difference, and d is component thickness.

[0018] The clustering algorithm in step S70 adopts the K-Means algorithm.

[0019] The temperature field simulation in step S30 uses the finite volume method for numerical calculation.

[0020] The sparse processing in step S40 adopts a method based on SVD decomposition.

[0021] The thermal bridge area identification in step S50 adopts an image segmentation algorithm based on the heat flux density gradient.

[0022] Wherein, the step S80 specifically includes:

[0023] S81. For the representative building components, select a representative building component that has an intersection with the multiple thermal bridge areas, and map its heat flux density value from the scale model with the smallest size ratio to a component corresponding to the scale model with a larger size ratio as an initial heat flux density estimate of the component at the larger size;

[0024] S82, using CUDA parallel computing capabilities on the GPU, for each scale model with a larger size ratio, taking the heat flux density estimate value of the scale model with the smallest size ratio as a starting point, performing local refinement calculation to obtain a refined heat flux density value of each component at the size ratio;

[0025] S83. Repeat S81 and S82 continuously, gradually expanding the size ratio until a civil engineering three-dimensional finite element model with a size of 1.00 times is reached, and the heat flux density of each building component in the non-thermal bridge area of ​​the model and the local heat flux density in the thermal bridge area are obtained.

[0026] The step S10 specifically includes:

[0027] Step 101: Obtain civil engineering design drawings and extract structural information and material parameters;

[0028] Step 102: establishing a three-dimensional finite element model based on the structural information and material parameters to reflect the physical geometric structure and material properties of the civil engineering project;

[0029] Step 103: Based on the three-dimensional finite element model, a series of scale models with size ratios ranging from 0.01 to 1.00 are generated, and the finite element mesh size of each model is equal. The purpose of this is to use differentiated calculation accuracy in the future and balance calculation accuracy and efficiency.

[0030] The step S20 specifically uses the CUDA parallel computing capability on the GPU to calculate the thermal performance parameters such as thermal resistance and heat transfer coefficient of each building component in the scale model with a minimum size ratio of 0.01. Since the GPU has powerful parallel computing capabilities, this step can efficiently complete the calculation of the thermal performance parameters of all components.

[0031] The step S30 specifically includes: still on the GPU, using the numerical simulation method, establishing the temperature field distribution of the scale model with the minimum size ratio of 0.01. Taking full advantage of the parallel computing advantage of the GPU, the temperature field of the scale model can be efficiently simulated.

[0032] The step S40 specifically includes:

[0033] Step 401: On the GPU, using the parallel algorithm provided by the Thrust library, the temperature field matrix is ​​sparsely processed to obtain a sparse matrix;

[0034] Step 402: Apply the Gaussian mixture model clustering algorithm to divide the sparse matrix into thermal bridge areas and non-thermal bridge areas according to the temperature distribution. The purpose of this step is to identify the thermal bridge areas in the model and lay the foundation for subsequent differential calculations.

[0035] The step S50 specifically includes: transmitting the clustering result obtained in step S40 back to the CPU, and identifying and extracting multiple thermal bridge areas in the three-dimensional model on the CPU. This step prepares for the subsequent calculation of the thermal bridge areas.

[0036] The step S60 is specifically: switching back to the GPU, and calculating the heat flux density of each building component in the non-thermal bridge area and the local heat flux density in the thermal bridge area according to the temperature field distribution inside and outside the building. The purpose of this step is to obtain thermal performance indicators of different areas and provide basic data for subsequent thermal insulation effect prediction.

[0037] The step S70 specifically includes:

[0038] Step 701: cluster analysis is performed on the thermal performance parameters of each building component of multiple scale models to obtain multiple cluster centers;

[0039] Step 702: Select the building component with the shortest distance from the cluster centers as the representative building component. The purpose of this step is to find a typical component that can represent the thermal performance of each component, providing a basis for subsequent mapping calculations.

[0040] The step S81 specifically includes:

[0041] Step 811, for the representative building components obtained in step S70, select those components that have intersections with multiple thermal bridge areas;

[0042] Step 812: Map the heat flux values ​​of these components in the model with the smallest scale of 0.01 times the model directly to the corresponding components of the model with a larger scale as the initial heat flux estimate of the component at the larger scale. This is done because the thermal bridge area requires more detailed calculations, and using the value of the smallest scale model as the initial estimate can provide a reasonable starting point for subsequent detailed calculations.

[0043] The step S82 specifically includes:

[0044] Step 821, using the CUDA parallel computing capability of the GPU for each model with a larger scale;

[0045] Step 822: Starting from the estimated heat flux value obtained in step S81, only the thermal bridge area determined in the model is finely simulated to obtain the fine heat flux value of each component under the size ratio. This can fully utilize the parallel advantage of GPU, improve the calculation efficiency, and ensure the calculation accuracy of the thermal bridge area.

[0046] The step S83 specifically includes:

[0047] Step 831, continuously repeating steps S81 and S82, gradually enlarging the size ratio until a three-dimensional finite element model with a size of 1.00 times is reached;

[0048] Step 832: In this process, the heat flux density values ​​of the components in the thermal bridge area in the minimum size model are first mapped to the larger size model as an initial estimate, and then a local refinement calculation is performed for the thermal bridge area of ​​each larger size model to obtain a refined heat flux density value. In this way, by gradually expanding the size, the heat flux density of each component in the non-thermal bridge area in the 1.00 times size model and the local heat flux density in the thermal bridge area can be finally obtained.

[0049] The training of the thermal insulation effect prediction model specifically includes:

[0050] Step 901, collect a large amount of data on geometric dimensions, material parameters, internal and external temperatures of actual construction projects;

[0051] Step 902: Calculate the thermal performance parameters such as heat flux density and thermal resistance of each component based on these data;

[0052] Step 903: using the thermal performance parameters as input features and the actually measured thermal insulation effects as labels, and using a supervised learning method to train a prediction model, such as a random forest, a neural network, etc., preferably a neural network;

[0053] Step 904: Adjust model complexity and hyperparameters to minimize prediction error and improve model generalization ability.

[0054] Optionally, the convergence threshold of the training of the prediction model may refer to industry standards such as heat flux density less than 5 W / ㎡, thermal resistance greater than 2.5 ㎡·K / W, etc.

[0055] Optionally, in step S81, the mapping of the heat flux density value adopts a linear interpolation method; optionally, the local refinement calculation in step S82 adopts a finite element method for numerical simulation; optionally, in step S83, the increment of gradually enlarging the size ratio is 0.2 times.

[0056] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for predicting thermal insulation effects in the civil engineering design stage.

[0057] A third aspect of the present invention provides a thermal insulation effect prediction system for a civil engineering design stage, which includes the above-mentioned computer-readable storage medium.

[0058] Compared with the prior art, the invention provides a method, medium and system for predicting the thermal insulation effect in the civil engineering design stage. The invention makes full use of the advantages of GPU parallel computing, introduces a multi-scale finite element model and differentiated calculation accuracy, and greatly improves the calculation efficiency while ensuring the calculation accuracy. The specific effects are as follows:

[0059] 1. Make full use of the parallel computing power of GPU to greatly improve computing efficiency: The method of the present invention makes full use of the powerful parallel computing power of GPU. In the key links such as calculating thermal performance parameters, establishing temperature field distribution, identifying thermal bridge areas and local refinement calculations, GPU parallel algorithms are used. Compared with CPU serial calculations, the calculation time can be greatly shortened and the computing efficiency can be improved.

[0060] 2. Introducing multi-scale finite element models to balance calculation accuracy and efficiency: The method of the present invention does not directly perform fine calculations on the entire three-dimensional finite element model, but generates a series of scale models with size ratios ranging from 0.01 times to 1.00 times. For the model with the smallest size of 0.01 times, a fine calculation method is used; for the model with larger size, a mapping and local refinement method based on the results of the smallest size model is used, which greatly reduces the amount of calculation while still ensuring the calculation accuracy of the thermal bridge area. This multi-scale modeling method effectively balances the calculation accuracy and efficiency while ensuring the overall calculation accuracy.

[0061] 3. Differentiated calculation accuracy, focusing on key thermal bridge areas: The method of the present invention fully considers the different characteristics of thermal bridge areas and non-thermal bridge areas in the thermal insulation effect analysis. For thermal bridge areas, due to the existence of abnormally high temperature gradients, more sophisticated local calculations are required; while for non-thermal bridge areas, moderately reducing the calculation accuracy has little effect on the final result. Therefore, the method of the present invention adopts differentiated calculation accuracy and focuses on key thermal bridge areas, which not only ensures the overall calculation accuracy, but also greatly improves the calculation efficiency.

[0062] 4. Integrate machine learning technology to improve prediction accuracy: The present invention adopts a method based on machine learning in the final thermal insulation effect prediction link. The thermal performance parameters of each component calculated in the above steps are used as input features, and the thermal insulation effect index can be output quickly and accurately using a pre-trained prediction model. Compared with traditional empirical formula predictions, this method of integrating machine learning can better capture the intrinsic relationship between complex thermal performance and thermal insulation effect, and improve the accuracy of prediction.

[0063] In general, the thermal insulation effect prediction method proposed in the present invention greatly improves the calculation efficiency while ensuring the calculation accuracy, and provides a strong technical support for the thermal insulation optimization in the civil engineering design stage. It solves the existing thermal insulation effect based on three-dimensional finite element analysis, which has the problem that if the entire model is directly refined, it will consume a lot of computing resources and be inefficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0065] Figure 1 A flow chart of the method provided by the present invention;

[0066] Figure 2 This is a specific flow chart of step S80. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the embodiments of the present invention more clear, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0068] like Figure 1 As shown, it is a flow chart of a method for predicting thermal insulation effect in the civil engineering design stage provided by the present invention, and the method comprises the following steps:

[0069] S10, establishing a civil engineering three-dimensional finite element model according to the structural information and material parameters of the civil engineering design drawings, and establishing multiple scale models with sequential size ratios based on the civil engineering three-dimensional finite element model, wherein the finite element mesh sizes of the multiple scale models are equal, and the sequential size ratio is a specified number of multiple size ratios from 0.01 times the size ratio to 1.00 times the size;

[0070] S20, using CUDA parallel computing capabilities to calculate thermal performance parameters of each building component of the scale model of the minimum size ratio on the GPU, including thermal resistance and heat transfer coefficient;

[0071] S30, continue to use the numerical simulation method on the GPU to establish the temperature field of the proportional model with the minimum size ratio;

[0072] S40, using the parallel algorithm provided by the Thrust library on the GPU to sparse the temperature field matrix to obtain a sparse matrix, and applying the Gaussian mixture model clustering algorithm to divide the sparse matrix into thermal bridge areas and non-thermal bridge areas according to the temperature distribution;

[0073] S50, transmitting the clustering result back to the CPU, identifying and extracting multiple thermal bridge areas existing in the 3D model;

[0074] S60, switching back to the GPU, and calculating the heat flux density of each building component in the non-thermal bridge area and the local heat flux density in the thermal bridge area according to the temperature field distribution inside and outside the building;

[0075] S70, clustering the thermal performance parameters of each building component of the multiple scale models to obtain multiple cluster centers, and obtaining multiple building components with the shortest distances to the cluster centers as representative building components;

[0076] S80, using representative building components, continuously mapping them to corresponding components of a more proportional scale model, and finally integrating them into a civil engineering three-dimensional finite element model of 1.00 times the size, to obtain the heat flux density of each building component in the non-thermal bridge area of ​​the civil engineering three-dimensional finite element model and the local heat flux density in the thermal bridge area;

[0077] S90. Finally, the heat flux density of each building component in the non-thermal bridge area of ​​the civil engineering three-dimensional finite element model and the local heat flux density in the thermal bridge area are transmitted back to the CPU, input into a pre-trained thermal insulation effect prediction model, and the thermal insulation effect index of the civil engineering design stage is obtained, and output to the designer.

[0078] Wherein, the step S80 specifically includes:

[0079] S81. For the representative building components, select a representative building component that has an intersection with the multiple thermal bridge areas, and map its heat flux density value from the scale model with the smallest size ratio to a component corresponding to the scale model with a larger size ratio as an initial heat flux density estimate of the component at the larger size;

[0080] S82, using CUDA parallel computing capabilities on the GPU, for each scale model with a larger size ratio, taking the heat flux density estimate value of the scale model with the smallest size ratio as a starting point, performing local refinement calculation to obtain a refined heat flux density value of each component at the size ratio;

[0081] S83. Repeat S80 and S81 continuously, gradually expanding the size ratio until a civil engineering three-dimensional finite element model with a size of 1.00 times is reached, and the heat flux density of each building component in the non-thermal bridge area of ​​the model and the local heat flux density in the thermal bridge area are obtained.

[0082] The specific implementation methods of the above steps are described in detail below:

[0083] The specific implementation method of step S10 is: first, obtain structural information and material parameters according to the civil engineering design drawings, and establish a corresponding three-dimensional finite element model. The three-dimensional finite element model reflects the physical geometric structure and material properties of the civil engineering project. Next, based on this three-dimensional finite element model, generate a series of multiple scale models of size ratios. The finite element mesh sizes of these scale models are equal, but the size ratio is gradually increased from 0.01 times to 1.00 times, and a specified number of intermediate size ratios are set. The purpose of doing this is to use calculation methods of different precisions in the future to balance calculation accuracy and efficiency.

[0084] The specific implementation of step S20 is: using the CUDA parallel computing capability on the GPU to calculate the thermal performance parameters of each building component in the scale model with a minimum size ratio of 0.01, including thermal resistance and heat transfer coefficient. Since the GPU has powerful parallel computing capabilities, this step can efficiently complete the calculation of the thermal performance parameters of all components.

[0085] The specific implementation of step S30 is: still on the GPU, using the method of numerical simulation, to establish the temperature field distribution of the scale model with the minimum size ratio of 0.01. This step can make full use of the parallel computing advantage of the GPU to efficiently simulate the temperature field of the scale model.

[0086] The specific implementation of step S40 is: on the GPU, the temperature field matrix is ​​sparsely processed using the parallel algorithm provided by the Thrust library to obtain a sparse matrix. Then, the Gaussian mixture model clustering algorithm is applied to divide the sparse matrix into thermal bridge areas and non-thermal bridge areas according to the temperature distribution. The purpose of this step is to identify the thermal bridge areas in the model and lay the foundation for subsequent differentiated calculations.

[0087] The specific implementation of step S50 is: the clustering result obtained in the previous step is transmitted back to the CPU, and multiple thermal bridge areas existing in the three-dimensional model are identified and extracted on the CPU. This step is to prepare for the subsequent thermal bridge area calculation.

[0088] The specific implementation of step S60 is: switch back to the GPU, and calculate the heat flux density of each building component in the non-thermal bridge area and the local heat flux density in the thermal bridge area according to the temperature field distribution inside and outside the building. The purpose of this step is to obtain thermal performance indicators in different areas and provide basic data for subsequent thermal insulation effect prediction.

[0089] The specific implementation of step S70 is: cluster analysis is performed on the thermal performance parameters of each building component of multiple scale models to obtain multiple cluster centers. Then, the building component with the shortest distance to these cluster centers is selected as the representative building component. The purpose of this step is to find a typical component that can represent the thermal performance of each component, providing a basis for subsequent mapping calculations.

[0090] The specific implementation method of step S80 is:

[0091] Sub-step S81: For the representative building components obtained in the previous step, select those components that intersect with multiple thermal bridge areas, and map their heat flux density values ​​from the model with the smallest scale of 0.01 times to the corresponding components of the model with a larger scale as the initial heat flux density estimate of the component at the larger scale.

[0092] Sub-step S82: Utilize the CUDA parallel computing capability on the GPU, and for each model with a larger scale, take the heat flux density estimate of the model with the smallest scale of 0.01 times as the starting point, perform local refinement calculations to obtain the refined heat flux density values ​​of each component at that scale.

[0093] Sub-step S83: Repeat sub-steps S81 and S82 continuously, gradually expanding the size ratio until a three-dimensional finite element model with a size of 1.00 times is reached, and the heat flux density of each building component in the non-thermal bridge area of ​​the model and the local heat flux density in the thermal bridge area are obtained.

[0094] The purpose of this step S80 is to gradually obtain the thermal performance parameters of each component in the 1.00 times size model by mapping and locally refining the heat flux value of the minimum size model, thereby ensuring the calculation accuracy and improving the calculation efficiency.

[0095] The specific implementation of step S90 is: the heat flux density of each building component in the non-thermal bridge area in the 1.00 times size model obtained in the previous step and the local heat flux density in the thermal bridge area are input into the pre-trained thermal insulation effect prediction model. The prediction model is based on a machine learning algorithm and can output the final thermal insulation effect index according to the input thermal performance parameters. Finally, this thermal insulation effect index is output to the designer to provide a reference for subsequent design optimization.

[0096] The specific implementation of step S81 is: for the representative building components obtained in the previous step S70, select those components that intersect with multiple thermal bridge areas, and directly map their heat flux values ​​from the model with the minimum scale of 0.01 times to the corresponding components of the model with a larger scale. The reason why the components that intersect with the thermal bridge area are selected here is that due to the existence of abnormal high temperature gradients in the thermal bridge area, more detailed local calculations and analyses are required. Mapping the heat flux value in the minimum scale model as an initial estimate to the larger scale model can provide a reasonable starting point for subsequent refinement calculations.

[0097] The specific implementation of step S82 is: using the CUDA parallel computing capability on the GPU, for each model with a larger scale, starting from the heat flux density estimation value obtained in step S81, perform local refinement calculation. Specifically, for each model with a larger scale, first determine the thermal bridge area therein, and then perform fine numerical simulation calculations only on these thermal bridge areas to obtain the fine heat flux density values ​​of each component under the scale. This can make full use of the parallel computing advantages of the GPU, improve the computing efficiency, and ensure the accuracy of the thermal bridge area calculation.

[0098] The specific implementation of step S83 is: repeating steps S81 and S82 continuously, gradually expanding the size ratio until a three-dimensional finite element model of 1.00 times the size is reached. In this process, the heat flux density value of the components in the thermal bridge area in the minimum size model is first mapped to the larger size model as an initial estimate, and then a local refinement calculation is performed for the thermal bridge area of ​​each larger size model to obtain a refined heat flux density value. In this way, by gradually expanding the size, the heat flux density of each component in the non-thermal bridge area in the 1.00 times size model and the local heat flux density in the thermal bridge area can be finally obtained.

[0099] The training process of the thermal insulation effect prediction model is as follows:

[0100] First, collect a large amount of relevant data from actual construction projects, including the geometric dimensions, material parameters, internal and external temperatures of the building. Then, calculate the thermal performance indicators of each component, such as heat flux density, thermal resistance, etc., based on these data. Next, use these thermal performance parameters as input features and the actual measured building insulation effects as labels to train the insulation effect prediction model using supervised learning methods. Commonly used algorithms include linear regression, neural networks, etc. During the training process, it is necessary to adjust the complexity and hyperparameters of the model to minimize the prediction error and improve the generalization ability of the model.

[0101] After training, the thermal insulation effect prediction model can be used in the actual civil engineering design stage. According to the thermal performance parameters of each component in the design scheme, the overall thermal insulation effect index can be quickly predicted to provide a basis for design optimization. The training threshold of the prediction model can refer to industry standards, such as heat flux less than 5W / m 2 , thermal resistance greater than 2.5m 2 K / W, etc.

[0102] In order to better understand the present method, a specific embodiment of the present method is described below in combination with a specific formula:

[0103] The specific implementation of step S10 is as follows:

[0104] First, the structural information and material parameters are obtained according to the civil engineering design drawings, and the corresponding three-dimensional finite element model is established. The three-dimensional finite element model can be expressed by the following formula:

[0105] M i (x, y, z) = f(G, M)

[0106] Among them, M i (x, y, z) represents the properties of the i-th unit in the three-dimensional finite element model, (x, y, z) is the spatial coordinate of the unit, G is the geometric structure information, and M is the material parameter. Function f represents the process of constructing a three-dimensional finite element model based on the geometric structure and material parameters.

[0107] Next, based on this 3D finite element model, a series of scale models with size ratios ranging from 0.01 to 1.00 are generated. The finite element mesh sizes of these scale models are equal, but the overall size changes. This process can be described by the following formula:

[0108] M i，j (x, y, z) = M i (s j x,s j y,s j z)

[0109] Among them, M i，j(x, y, z) represents the i-th unit attribute at the j-th scale, s j is the jth size ratio, with a value range of [0.01, 1.00], and a total of N intermediate ratio values ​​are set.

[0110] The purpose of generating these scale models is to use different calculation methods with different degrees of precision in the future to balance the calculation accuracy and efficiency.

[0111] The specific implementation of step S20 is as follows:

[0112] Using the CUDA parallel computing capability on the GPU, the thermal performance parameters of each building component in the scale model with the minimum scale ratio s1 = 0.01 are calculated, including the thermal resistance R and the heat transfer coefficient h. It can be described by the following formula:

[0113]

[0114] Among them, d i is the thickness of the i-th component, λ i As GPU has powerful parallel computing capability, this step can efficiently complete the calculation of thermal performance parameters of all components.

[0115] The specific implementation of step S30 is as follows:

[0116] Still on the GPU, using the numerical simulation method, the temperature field distribution of the scale model with the minimum scale s1 = 0.01 is established. It can be described by the following heat conduction equation:

[0117]

[0118] Among them, ρ is density, c is specific heat capacity, T is temperature, t is time, λ is thermal conductivity, and Q is internal heat source term. By solving the partial differential equation, the transient temperature field distribution of the scale model can be obtained. This step can make full use of the parallel computing advantages of GPU to efficiently simulate the temperature field of the scale model.

[0119] The specific implementation of step S40 is as follows:

[0120] On the GPU, the temperature field matrix is ​​sparsely processed using the parallel algorithm provided by the Thrust library to obtain a sparse matrix T sparse . It can be described by the following formula:

[0121] T sparse = Sparsify(T)

[0122] Where T is the original temperature field matrix, and Sparsify() represents the sparsification function.

[0123] Next, the Gaussian mixture model clustering algorithm is applied to cluster the sparse matrix T according to the temperature distribution. sparse It is divided into thermal bridge area and non-thermal bridge area. The Gaussian mixture model can be expressed as:

[0124]

[0125] Among them, x is the observed data (temperature value), θ=(π k , μ k ,∑ k} is the model parameter, K is the number of Gaussian components. The model parameters are iteratively optimized by the EM algorithm, and T sparse The purpose of this step is to identify the thermal bridge areas in the model and lay the foundation for subsequent differentiation calculations.

[0126] The specific implementation of step S50 is as follows:

[0127] The clustering results obtained in the previous step are transmitted back to the CPU, where multiple thermal bridge areas existing in the 3D model are identified and extracted. and heat flux threshold qt h The method for regional division is:

[0128]

[0129] Among them, R bridge Represents the thermal bridge area, and x is the spatial coordinate of the unit. This step is to prepare for the subsequent thermal bridge area calculation.

[0130] The specific implementation of step S60 is as follows:

[0131] Switch back to the GPU and calculate the heat flux density qn of each building component in the non-thermal bridge area based on the temperature field distribution inside and outside the building. b , and the local heat flux density q in the thermal bridge area b .

[0132] For non-thermal bridge areas, the heat flux can be calculated using the following formula:

[0133]

[0134] Where, ΔT i is the temperature difference of the components in the i-th non-thermal bridge area.

[0135] For the thermal bridge area, due to the complex three-dimensional temperature gradient, the heat flux density needs to be calculated by integration:

[0136]

[0137] Among them, A i is the surface area of ​​the component in the ith thermal bridge region, is its boundary and n is the unit normal vector.

[0138] The purpose of this step is to obtain the thermal performance indicators of different areas and provide basic data for subsequent insulation effect prediction.

[0139] The specific implementation of step S70 is as follows:

[0140] Cluster analysis is performed on the thermal performance parameters of each building component of multiple scale models, such as thermal resistance R and heat transfer coefficient h, to obtain K cluster centers K-Means algorithm can be used for clustering:

[0141]

[0142] Among them, xi is the thermal performance parameter vector of the i-th component.

[0143] Then, the building components with the shortest distance to these cluster centers are selected as representative building components. The purpose of this step is to find typical components that can represent the thermal performance of each component and provide a basis for subsequent mapping calculations.

[0144] The specific implementation of step S80 is as follows:

[0145] Sub-step S81: For the representative building components obtained in the previous step, select those components that intersect with multiple thermal bridge areas and calculate their heat flux values ​​q b，rep The model with the smallest scale s1 is directly mapped to the larger scale s j The model corresponds to the component, as the initial heat flux estimate of the component at a larger size

[0146]

[0147] The reason why we choose the components that intersect with the thermal bridge area is that the thermal bridge area needs more detailed local calculation and analysis due to the abnormal high temperature gradient. Mapping the heat flux density value in the smallest size model as the initial estimate to the larger size model can provide a reasonable starting point for subsequent detailed calculations.

[0148] Sub-step S82: Using the CUDA parallel computing capability on the GPU, for each larger size ratio s j The model is based on the estimated heat flux obtained in step S81. As the starting point, only the thermal bridge area R bridgePerform detailed numerical simulation to obtain the detailed heat flux value q of each component under this size ratio. b，i，j :

[0149]

[0150] Doing so can fully utilize the parallel computing advantages of the GPU, improve computing efficiency, and ensure the accuracy of thermal bridge area calculations.

[0151] Sub-step S83: Repeat steps S81 and S82 continuously, gradually enlarging the size ratio s j , until a 3D finite element model with a size of 1.00 is reached. In this process, the heat flux density values ​​of the components in the thermal bridge area of ​​the smallest size model are first mapped to the larger size model as the initial estimate, and then local refinement calculations are performed on the thermal bridge area of ​​each larger size model to obtain a refined heat flux density value. In this way, by gradually expanding the size, the heat flux density q of each component in the non-thermal bridge area of ​​the 1.00 size model can be obtained. nb，i，N , and the local heat flux density q in the thermal bridge area b，i，N .

[0152] The specific implementation of step S90 is as follows:

[0153] The heat flux density q of each building component in the non-thermal bridge area of ​​the 1.00 times scale model obtained in the previous step is nb，i，N and the local heat flux density q in the thermal bridge area b，i，N , is input into the pre-trained thermal insulation effect prediction model. The prediction model can be expressed as follows:

[0154] y=f(q nb，1，N， q nb，2，N ,…,q nb，M，N ,q b，1，N ,q b，2，N ,…,q b，L，N )

[0155] Among them, y is the insulation effect index, M and L are the number of components in the non-thermal bridge area and the thermal bridge area respectively, and f is the prediction model, which can use machine learning algorithms such as linear regression and neural network.

[0156] Finally, the thermal insulation effect index y is output to the designers to provide a reference for subsequent design optimization.

[0157] In summary, this thermal insulation effect evaluation method based on multi-scale finite element model, differentiated calculation accuracy and machine learning prediction makes full use of the advantages of GPU parallel computing. While ensuring the calculation accuracy, it greatly improves the calculation efficiency and provides effective technical support for thermal insulation optimization in the civil engineering design stage.

[0158] The relevant variables of the above embodiment are explained as follows: i (x, y, z): the properties of the i-th unit in the three-dimensional finite element model, including geometric dimensions, material parameters, etc.; G: geometric structure information of civil engineering; M: parameters of materials used in civil engineering, such as thermal conductivity, specific heat capacity, etc.; M i，j (x, y, z): the i-th unit attribute at the j-th scale; s j : jth size ratio, the value range is [0.01, 1.00]; N: the number of intermediate size ratios to be set; R i，1 : thermal resistance of the ith component in the minimum scale model; h i，1 : heat transfer coefficient of the ith component in the minimum size model; d i : the thickness of the i-th component; λ i : thermal conductivity of the i-th component; ρ: material density; c: material specific heat capacity; T: temperature; t: time; Q: internal heat source term; T sparse : The sparse matrix obtained after the temperature field matrix is ​​sparsely processed; x: observed data (temperature value); θ: parameters of the Gaussian mixture model, including the mixing coefficient π k , mean μ k and covariance∑ k ; K: number of Gaussian components; Temperature gradient; Temperature gradient threshold; q: heat flux density; q th : Heat flux density threshold; R bridge : Thermal bridge area; q nb，i : heat flux density of the component in the i-th non-thermal bridge area; q b，i : local heat flux density of the components in the ith thermal bridge area; A i : the surface area of ​​the i-th component; The boundary of the i-th component; n: unit normal vector; X i : thermal performance parameter vector of the i-th component; c k : kth cluster center; x rep，k : The representative building component corresponding to the kth cluster center; q b，rep : Local heat flux density of representative building components in the minimum size model; The estimated initial heat flux density of the i-th component in the j-th scale model; q b，i，j : The fine heat flux value of the i-th component in the j-th size model; q nb，i，N : heat flux density of the i-th non-thermal bridge area component in the 1.00 times scale model; q b，i，N: local heat flux density of the i-th thermal bridge area component in the 1.00 times scale model; y: thermal insulation effect index; M: total number of components in the non-thermal bridge area; L: total number of components in the thermal bridge area; f: thermal insulation effect prediction model; i: unit or component number; j: size ratio number; k: cluster center number; nb: non-thermal bridge area; b: thermal bridge area; rep: representative building component; Temperature gradient threshold; q th : Heat flux density threshold.

[0159] A second aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the above-mentioned method for predicting thermal insulation effects in the civil engineering design stage.

[0160] A third aspect of the present invention provides a thermal insulation effect prediction system for a civil engineering design stage, which includes the above-mentioned computer-readable storage medium.

[0161] Specifically, the principle of the present invention is: currently, the smaller the size of the civil engineering finite element grid, the higher the calculation accuracy, but at the same time, the more hardware resources are consumed, and the calculation time will be correspondingly longer. For large-scale civil engineering projects, their three-dimensional finite element models usually have a huge number of grids. If the entire model is directly refined, it will consume a lot of computing resources and be inefficient. At the same time, in the thermal insulation effect analysis, the thermal bridge area requires more refined local calculations and analysis due to the existence of abnormally high temperature gradients, while for non-thermal bridge areas, moderately reducing the accuracy has little effect on the final result. Therefore, how to use different degrees of precision for different areas under limited computing resources to balance calculation accuracy and efficiency is an important problem that needs to be solved in this technical field. The specific principles of the technical solution adopted by the present invention are as follows:

[0162] 1. Multi-scale finite element model: In order to balance the calculation accuracy and efficiency, the method of the present invention does not directly perform fine calculations on the entire three-dimensional finite element model, but generates a series of proportional models with different size ratios based on the original model. The finite element mesh sizes of these proportional models are equal, only the overall geometric dimensions change. The purpose of this is to use a fine calculation method for the model with the smallest size ratio; while for the larger size model, a mapping and local refinement method based on the results of the smallest size model is used, which greatly reduces the amount of calculation. Through this multi-scale modeling method, both the overall calculation accuracy is guaranteed and the calculation efficiency is improved.

[0163] 2. Differentiated calculation accuracy: In the analysis of thermal insulation effect, the thermal bridge area needs more refined local calculation and analysis due to the existence of abnormally high temperature gradients; while for the non-thermal bridge area, moderately reducing the accuracy has little effect on the final result. Therefore, the method of the present invention adopts a differentiated calculation accuracy strategy, performs fine calculations on the thermal bridge area, and moderately reduces the calculation accuracy for the non-thermal bridge area. Specifically, the GPU parallel algorithm is first used to quickly identify the thermal bridge area in the three-dimensional model, and then only these thermal bridge areas are subjected to fine numerical simulation calculations, while for the non-thermal bridge area, a mapping method based on the minimum size model results is adopted. This not only ensures the accuracy of the thermal bridge area calculation, but also greatly improves the overall calculation efficiency.

[0164] 3. GPU parallel computing: The method of the present invention makes full use of the powerful parallel computing capabilities of GPU. GPU acceleration is used to calculate thermal performance parameters, establish temperature field distribution, identify thermal bridge areas, and perform local refinement calculations. Compared with CPU serial calculations, GPU parallel algorithms can greatly shorten the calculation time and greatly improve the calculation efficiency. This design idea that fully utilizes the advantages of GPU is the key to the efficient implementation of the method of the present invention.

[0165] 4. Machine learning fusion: In the final thermal insulation effect prediction link, the method of the present invention adopts a method based on machine learning. The thermal performance parameters of each component calculated in the above steps are used as input features, and the thermal insulation effect index can be output quickly and accurately using a pre-trained prediction model. Compared with traditional empirical formula predictions, this method of integrating machine learning can better capture the intrinsic relationship between complex thermal performance and thermal insulation effect, and improve the accuracy of prediction.

[0166] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

[0167] In order to better understand the present invention, a specific embodiment is now designed: the thermal insulation effect prediction method proposed in the present invention has been applied to the design optimization process of a large residential project. The project is a 24-story residential building with a construction area of ​​about 32,000 square meters. Taking this project as an example, the implementation process of the method of the present invention and its technical effect are specifically introduced.

[0168] First, based on the architectural design drawings of the residential building, its structural information and material parameters were extracted, and the corresponding three-dimensional finite element model was established. The model contains detailed information such as the building's geometric dimensions, exterior wall materials, and window locations. The number of model grids is about 2 million.

[0169] Based on this 3D finite element model, a series of scale models with size ratios ranging from 0.01 to 1.00 were generated. Specifically, six intermediate scale ratios were set: 0.01, 0.20, 0.40, 0.60, 0.80 and 1.00. The finite element mesh sizes of these scale models are equal, only the overall geometric dimensions change.

[0170] Next, the CUDA parallel computing capability on the GPU was used to calculate the thermal performance parameters of each building component in these scale models, including thermal resistance and heat transfer coefficient. Taking the scale model with the smallest size of 0.01 as an example, the thermal resistance R of the i-th component i，1 and the heat transfer coefficient h i，1 It can be calculated using the following formula:

[0171]

[0172] Among them, d i is the thickness of the i-th component, λ i The thermal performance parameters of all 2 million components were calculated within 10 seconds using GPU parallel computing.

[0173] Next, still on the GPU, the temperature field distribution of the scale model with a minimum size of 0.01 was established using numerical simulation methods. Specifically, the following heat conduction equation was solved:

[0174]

[0175] Among them, ρ is density, c is specific heat capacity, T is temperature, t is time, λ is thermal conductivity, and Q is internal heat source term. By iteratively solving the partial differential equation, the transient temperature field distribution of the scale model was obtained within 20 seconds.

[0176] With the temperature field distribution, we then use the parallel algorithm provided by the Thrust library to perform sparse processing on the temperature field matrix and obtain the sparse matrix T sparse Then, the Gaussian mixture model clustering algorithm is applied to cluster T according to the temperature distribution. sparse It is divided into thermal bridge area and non-thermal bridge area. The mathematical model of clustering algorithm can be expressed as:

[0177]

[0178] Among them, x is the observed data (temperature value), θ=(π k , μ k ,∑ k} is the model parameter, K is the number of Gaussian components. The model parameters are iteratively optimized by the EM algorithm, and T sparseThe system was successfully divided into four clusters, two of which were identified as thermal bridge areas. The entire clustering process took only 15 seconds.

[0179] The clustering results are sent back to the CPU, where the thermal bridge areas in the 3D model are further extracted and analyzed. Specifically, according to the temperature gradient threshold and heat flux threshold qt h =10W / m 2 , the area in the temperature field distribution that meets the following conditions is divided into the thermal bridge area R bridge :

[0180]

[0181] Through this step, three main thermal bridge areas were successfully extracted, which were located around windows, at balcony connections, and at some wall junctions.

[0182] Next, we switch back to the GPU and calculate the heat flux density q of each building component in the non-thermal bridge area based on the temperature field distribution inside and outside the building. nb，i , and the local heat flux density q in the thermal bridge area b，i For non-thermal bridge areas, the heat flux can be calculated using the following formula:

[0183]

[0184] Where, ΔT i is the temperature difference of the components in the i-th non-thermal bridge area. For the thermal bridge area, due to the complex three-dimensional temperature gradient, the heat flux density needs to be calculated by integration:

[0185]

[0186] Among them, A i is the surface area of ​​the component in the ith thermal bridge region, is its boundary, and n is the unit normal vector. Using the parallel computing capability of GPU, the heat flux density calculation of all components was completed within 30 seconds.

[0187] With the thermal performance parameters of each component, cluster analysis was performed on multiple scale models, and 6 cluster centers were obtained. Then, the building components with the shortest distance to these cluster centers are selected as representative building components.

[0188] Based on these representative building components, the mapping and local refinement calculation in step S80 are implemented. First, representative components that intersect with multiple thermal bridge areas are selected and their local heat flux density q in the minimum size 0.01 times model is calculated. b，repDirectly map to the corresponding component of the model at a larger scale as an initial estimate

[0189]

[0190] Next, using the CUDA parallel computing capability of the GPU, for each larger scale model, starting from the above initial estimate, only the determined thermal bridge area R bridge Perform detailed numerical simulation to obtain the detailed heat flux density value q of each component under this size. b，i，j :

[0191]

[0192] This mapping and local refinement process was iterated repeatedly, and the scale ratio was gradually enlarged. Finally, the thermal performance parameters of each component in the 1.00 times scale model were obtained, including the heat flux density q in the non-thermal bridge area. nb，i，6 and the local heat flux density q in the thermal bridge area b，i，6 The entire mapping and refinement calculation process took 2 minutes.

[0193] With the thermal performance parameters of the above components, they are finally input into the pre-trained thermal insulation effect prediction model to output the thermal insulation effect index of the residential building. The prediction model uses a regression algorithm based on random forests, and its mathematical form is:

[0194] y=f(q nb，1，6 ,q nb，2，6 , ..., q nb，M，6 ,q b，1，6 ,q b，2，6 ,......,q b，L，6 )

[0195] Among them, y is the thermal insulation effect index, M and L are the number of components in the non-thermal bridge area and the thermal bridge area respectively, and f is the prediction model. Through machine learning training, the model can output an index that comprehensively reflects the overall thermal insulation effect based on the input thermal performance parameters of each component.

[0196] After applying the method of the present invention to predict the insulation effect of the residential building project, the results obtained are as follows: the overall heat loss coefficient is 0.35W / (m 2 ·K), the overall heat transfer coefficient is 2.85W / (m 2 ·K), the overall thermal resistance is 0.35(m 2 These indicators show that the thermal insulation performance of the residential building is good and can meet the requirements of the national building energy conservation standards.

[0197] The above prediction results were compared with the method based on traditional three-dimensional finite element analysis. The traditional method directly performs fine calculations on the entire three-dimensional finite element model, which takes about 2 hours. However, the method of the present invention, while ensuring the calculation accuracy, completes the calculation in only 2 minutes by introducing multi-scale modeling and differentiated calculation accuracy, greatly improving the calculation efficiency, which is about 60 times that of the traditional method.

[0198] The performance of the method of the present invention in each calculation step is further analyzed:

[0199] 1. Calculation of thermal performance parameters: Using GPU parallel algorithm, the calculation of all 2 million components was completed in just 10 seconds, which is 40 times faster than the CPU serial method.

[0200] 2. Temperature field distribution modeling: GPU numerical simulation was used to obtain the temperature field of the smallest size model within 20 seconds, which is 60 times faster than the CPU method.

[0201] 3. Thermal bridge area identification: Using GPU parallel clustering algorithm, the thermal bridge area was divided within 15 seconds, which is 80 times faster than the CPU method.

[0202] 4. Heat flux calculation: GPU parallel computing was used to complete the heat flux calculation of all components within 30 seconds, which is 50 times faster than the CPU method.

[0203] 5. Mapping and local refinement: Detailed calculations are performed on the thermal bridge area, while mapping methods are used for the non-thermal bridge area. The entire process takes only 2 minutes, which is 60 times faster than detailed calculations of the entire model.

[0204] In summary, the method of the present invention, while ensuring the calculation accuracy, greatly improves the calculation efficiency by making full use of the parallel computing power of GPU, introducing multi-scale modeling and differentiated calculation accuracy, and integrating machine learning technology, providing strong support for the thermal insulation optimization in the civil engineering design stage. Taking the above-mentioned residential building project as an example, the calculation time of the method of the present invention is shortened by 60 times compared with the traditional three-dimensional finite element analysis method, which fully demonstrates its superior computing performance.

[0205] In addition, the reliability of the method of the present invention in terms of calculation accuracy was verified through comparative analysis. Several representative components were selected and the thermal performance parameters calculated by the method of the present invention and the traditional method were compared. The results showed that the deviation of the two methods was within 5%, which met the accuracy requirements of engineering applications. This shows that the method of the present invention can greatly improve the calculation efficiency while ensuring the calculation accuracy.

[0206] In summary, the thermal insulation effect prediction method proposed in the present invention fully integrates cutting-edge technologies such as GPU parallel computing, multi-scale modeling, differentiated computing accuracy, and machine learning, providing strong support for thermal insulation optimization in the civil engineering design stage. Taking the above-mentioned residential building project as an example, the method of the present invention shortens the calculation time by 60 times compared with the traditional three-dimensional finite element analysis method, and can also meet the calculation accuracy requirements required for engineering applications.

Claims

1. A method for predicting thermal insulation effect in the civil engineering design stage, characterized in that: The following steps are involved: S10, establishing a civil engineering three-dimensional finite element model according to the structural information and material parameters of the civil engineering design drawings, and establishing multiple scale models of sequential size ratios based on the civil engineering three-dimensional finite element model, wherein the finite element mesh sizes of the multiple scale models are equal, and the sequential size ratio is a specified number of multiple size ratios from 0.01 times the size ratio to 1.00 times the size; S20, using CUDA parallel computing capabilities, calculate the thermal performance parameters of each building component of the scale model of the minimum size ratio on the GPU, including thermal resistance and heat transfer coefficient; S30, continue to use the numerical simulation method on the GPU to establish the temperature field of the proportional model with the minimum size ratio; S40, sparseizing the temperature field matrix to obtain a sparse matrix, and dividing the sparse matrix into thermal bridge areas and non-thermal bridge areas according to temperature distribution; S50, transmitting the clustering result back to the CPU, identifying and extracting multiple thermal bridge areas existing in the 3D model; S60, switch back to the GPU, and calculate the heat flux density of each building component in the non-thermal bridge area and the local heat flux density in the thermal bridge area according to the temperature field distribution inside and outside the building; S70, clustering the thermal performance parameters of each building component of the multiple scale models to obtain multiple cluster centers, and obtaining multiple building components with the shortest distances to the cluster centers as representative building components; S80, using representative building components, continuously mapping them to corresponding components of a more proportional scale model, and finally integrating them into a civil engineering three-dimensional finite element model of 1.00 times the size, to obtain the heat flux density of each building component in the non-thermal bridge area of ​​the civil engineering three-dimensional finite element model and the local heat flux density in the thermal bridge area; S90, finally, the heat flux density of each building component in the non-thermal bridge area of ​​the civil engineering three-dimensional finite element model and the local heat flux density in the thermal bridge area are transmitted back to the CPU, input into the pre-trained thermal insulation effect prediction model, obtain the thermal insulation effect index in the civil engineering design stage, and output to the designer; Wherein, the step S80 specifically includes: S81. For the representative building components, select a representative building component that has an intersection with the multiple thermal bridge areas, and map its heat flux density value from the scale model with the smallest size ratio to a component corresponding to the scale model with a larger size ratio as an initial heat flux density estimate of the component at the larger size; S82, using CUDA parallel computing capabilities on the GPU, for each scale model with a larger size ratio, taking the heat flux density estimate value of the scale model with the smallest size ratio as a starting point, performing local refinement calculations to obtain the refined heat flux density values ​​of each component at the size ratio; S83. Repeat S81 and S82 continuously, gradually expanding the size ratio until the civil engineering three-dimensional finite element model reaches 1.00 times the size, and obtain the local heat flux density in the thermal bridge area; for the heat flux density of each building structure in the non-thermal bridge area, a mapping method based on the minimum size model result is used to obtain it.

2. A method for predicting thermal insulation effect in the civil engineering design stage according to claim 1, characterized in that: The method of dividing the sparse matrix into thermal bridge areas and non-thermal bridge areas according to the temperature distribution is the Gaussian mixture model clustering algorithm.

3. The method for predicting thermal insulation effect in the civil engineering design stage according to claim 1, characterized in that: The formula for calculating heat flux density is: q = λ * ΔT / d, where q is the heat flux density, λ is the thermal conductivity of the material, ΔT is the temperature difference, and d is the thickness of the component.

4. The method for predicting thermal insulation effect in the civil engineering design stage according to claim 1, characterized in that: The clustering algorithm in step S70 adopts the K-Means algorithm.

5. The method for predicting thermal insulation effect in the civil engineering design stage according to claim 1, characterized in that: The temperature field simulation in step S30 uses the finite volume method for numerical calculation.

6. The method for predicting thermal insulation effect in the civil engineering design stage according to claim 1, characterized in that: The sparse processing in step S40 adopts a method based on SVD decomposition.

7. The method for predicting thermal insulation effect in the civil engineering design stage according to claim 1, characterized in that: The thermal bridge area identification in step S50 adopts an image segmentation algorithm based on the heat flux density gradient.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores program instructions, and when the program instructions are executed, they are used to execute the thermal insulation effect prediction method in the civil engineering design stage according to any one of claims 1 to 7.

9. A thermal insulation effect prediction system for the civil engineering design stage, characterized in that: Contains the computer-readable storage medium of claim 8.

Images