An airbag template ice shell building weatherability optimization design method based on an agent model

Abstract

The application discloses a kind of based on agent model's air bag template ice shell building weatherability optimization design method, it is related to building design technical field.The problems that existing technology lacks the quantitative evaluation index of ice shell building weatherability and lacks the construction of universal agent model are solved, including: S1 stage is established microclimate response agent model, through orthogonal experimental design to generate ice shell case set, extract morphological characteristics and spatial characteristics, obtain wind snow distribution label by combining CFD simulation, construct neural network agent model;S2 stage is fused multiple tools to realize performance optimization, utilize Grasshopper parametric modeling to extract features, call agent model to predict surface snow wind distribution, obtain radiation data jointly by Ladybug, combine sublimation rate to calculate initial and later strain energy and weatherability index, finally, morphological parameters are multi-objective optimized by NSGA-II algorithm, it has application prospect in the field of ice shell building structure optimization.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of architectural design, and particularly relates to a kind of air bag template ice shell building weatherability optimization design method based on agent model. BACKGROUND

[0002] At present, the ice shell building weatherability evaluation technology is still in the blank stage, and the core difficulty comes from the complexity of the performance degradation mechanism of the composite ice material under long-term climate action. Although existing research has revealed the sublimation rate calculation model of ice material and the dynamic change rule of elastic modulus, a systematic quantitative evaluation method of weatherability has not yet been formed, resulting in a lack of operational index basis for weatherability optimization design. At the same time, the agent model technology in the field of architectural design has formed a relatively standardized application paradigm: based on Latin hypercube sampling to extract feature-label data set, through artificial neural network (ANN), convolutional neural network (CNN) and other algorithms to train black box model, to replace the traditional simulation process to realize efficiency improvement. However, the above related researches have significant limitations: first, existing researches mostly focus on traditional performance targets such as form and structure, and the model construction for weatherability prediction is completely missing, which is rooted in the blank of quantitative index of ice shell building weatherability; second, the existing model highly depends on the design parameters and task boundaries defined by the designer subjectively, resulting in poor model generalization, repeated sampling and training when the project is replaced, the efficiency advantage is offset by the secondary modeling cost, and it cannot meet the cross-project universality demand. The above technical breakpoints lead to the dual bottlenecks of uncontrollable weatherability and low multi-objective coordination efficiency in the performance optimization of ice shell building. SUMMARY

[0003] To solve the problems of lack of quantitative evaluation index of ice shell building weatherability and lack of agent model construction with universality in the prior art, the present application provides an air bag template ice shell building weatherability optimization design method based on agent model, comprising:

[0004] S1, establishing an agent model of spatial form features and microclimate response;

[0005] S11, constructing an ice shell building case set through orthogonal experiment design, extracting form features and spatial features;

[0006] The form features include Z-axis slope, normalized rib distance, normalized height, maximum principal curvature and minimum principal curvature; and the spatial features include Y-axis slope and normalized wind direction distance.

[0007] S12, performing CFD simulation of wind environment and snow distribution on the ice shell building case set to obtain label information;

[0008] S13, model training is performed based on the morphological features, spatial features and label information, an artificial neural network proxy model is constructed, and the artificial neural network comprises: 1 input layer, 4 hidden layers and 1 output layer;

[0009] S2, weather resistance index calculation and multi-objective optimization are performed;

[0010] S21, the weather resistance index of the ice shell building is defined;

[0011] S22, a Grasshopper platform is used to construct an ice shell building parameterized model, and the morphological features and the spatial features are extracted;

[0012] S23, the proxy model in S1 is called back through a gh_Cpython plug-in, the morphological features and the spatial features are input, and snow distribution and wind environment data of the ice shell surface are obtained; surface radiation distribution data are obtained in combination with a Ladybug plug-in;

[0013] S24, based on the snow distribution and wind environment data and the surface radiation distribution data of the ice shell surface, morphological construction information, load information and material performance information of the ice shell building are updated and obtained, and a sublimation rate is calculated;

[0014] S25, based on the morphological construction information, load information and material performance information of the ice shell building and the sublimation rate, strain energy in the initial service period and the later service period is calculated through a Karamba3D plug-in, and then the weather resistance index of the ice shell building is calculated;

[0015] S26, the initial strain energy and the weather resistance index are taken as targets, and parameter optimization is realized through an NSGA-II algorithm of a Wallacei plug-in.

[0016] Further, in S11, in the morphological features, the Z-axis slope is obtained by:

[0017]

[0018] wherein, S Z is the Z-axis slope, N is the number of neighborhood unit surfaces, is a unit normal vector, is a unit vector in the Z-axis direction;

[0019] The normalized rib line distance is obtained by:

[0020]

[0021] wherein, is the normalized rib line distance, D j,i is the distance from point j to rib line i, D jis all the shortest distance data;

[0022] The normalized height is obtained by:

[0023]

[0024] wherein H z is the z coordinate data of the point, H j is the z coordinate data of all points;

[0025] The maximum principal curvature and the minimum principal curvature are obtained by:

[0026]

[0027] wherein C(v) is the principal curvature tensor, |B| is the area of the vertex 1-ring neighborhood, β(e) is the included angle of the normal vectors on both sides of the edge e, and |e∩B| is the length of the edge e in the neighborhood B, is the tensor product of the edge direction.

[0028] Further, in S11, in the spatial features, the Y-axis slope is obtained by:

[0029]

[0030] wherein S y is the Y-axis slope, is the unit vector in the y-axis direction;

[0031] The normalized wind direction distance is obtained by:

[0032]

[0033] wherein D is the normalized wind direction distance, D w is the vertical distance data of the point to the wind inflow interface, D n is the distance data of all points.

[0034] Further, in S13, the artificial neural network agent model is constructed, specifically: for snow distribution prediction, a 4-factor model containing the Z-axis slope, the normalized height, the distance from the rib line, and the minimum principal curvature is screened out; for wind environment prediction, a 7-factor model containing the Z-axis slope, the Y-axis slope, the normalized height, the distance from the rib line, the distance from the wind direction, the maximum principal curvature, and the minimum principal curvature is screened out.

[0035] Further, in S21, the weather resistance index of the ice rind building is obtained by:

[0036] W = (U E -U S ) / US

[0037] is obtained, wherein W is the weather resistance index of the ice shell building, U E is the elastic energy at the initial stage, U S is the elastic energy at a specified service time;

[0038] The elastic energy is obtained by:

[0039]

[0040] is obtained, wherein F i is the load of the node, μ i is the displacement of the node.

[0041] Further, in S24, the sublimation rate is obtained by:

[0042]

[0043] T s = T a + 0.0559q G

[0044] is obtained, wherein E FRI is the sublimation rate, v is the wind speed, ω is the paper fiber concentration, H is the cumulative sublimation height, T s is the material surface temperature, T a is the air temperature, q G is the solar radiation intensity.

[0045] Advantages of the present application:

[0046] Firstly, compared with the prior art, the present application can establish a general agent model for the air bag template ice shell building. The input independent variable is not affected by the designer's subjective intention, and is directly extracted from the generated building shape and spatial position. Therefore, when facing the air bag template ice shell building engineering project, the pre-trained agent model can be selected according to the building type. It is not necessary to repeat the training process, and the efficiency of the design process is truly improved.

[0047] Secondly, compared with the prior art, the present application first proposes a quantitative evaluation index of the weather resistance of the ice shell building. The multi-objective optimization technology in building design can be implemented in the weather resistance design of the ice shell building. This is never involved in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0048] Figure 1 is the flow chart of the design system of the present application;

[0049] Figure 2 is the integrated technology roadmap of the present application;

[0050] Figure 3 Ice rind multi-objective optimization morphology-performance trade-off comparison chart;

[0051] Figure 4 Model training index comparison chart for snow distribution label;

[0052] Figure 5 Model morphology verification effect chart for snow distribution label;

[0053] Figure 6 Model training index comparison chart for wind environment label;

[0054] Figure 7 Model training index comparison chart for wind environment label. DETAILED DESCRIPTION

[0055] The technical solutions of the present application will be further described below in combination with embodiments, but are not limited thereto. Any modification or equivalent replacement to the technical solutions of the present application without departing from the spirit and scope of the present application shall be covered in the protection scope of the present application. The process equipment or device not specifically mentioned in the following examples is the conventional equipment or device in the art. If not specifically mentioned, the raw materials used in the examples of the present application are commercially available. If not specifically mentioned, the technical means used in the examples of the present application is the conventional means known to those skilled in the art.

[0056] Example 1, a kind of airbag template ice rind building weather resistance optimization design method based on agent model, comprising:

[0057] S1, the establishment of space morphology characteristics and microclimate response agent model;

[0058] S11, through orthogonal experimental design to build ice rind building case set, extract morphology characteristics and space characteristics;

[0059] The morphology characteristics include: Z-axis slope, normalized rib line distance, normalized height, maximum principal curvature and minimum principal curvature;The space characteristics include: Y-axis slope and normalized wind direction distance;

[0060] S12, the ice rind building case set is subjected to wind environment and snow distribution CFD simulation, and label information is obtained;

[0061] S13, model training is carried out based on the morphology characteristics, space characteristics and label information, an artificial neural network agent model is constructed, and the artificial neural network includes: 1 input layer, 4 hidden layers and 1 output layer;

[0062] S2, weather resistance index calculation and multi-objective optimization are carried out;

[0063] S21, define the weather resistance index of ice shell building;

[0064] S22, build an ice shell building parameterized model using the Grasshopper platform, and extract the morphological features and spatial features;

[0065] S23, call back the agent model in S1 through the gh_Cpython plug-in, input the morphological features and spatial features, and obtain snow distribution and wind environment data of the ice shell surface; combine the Ladybug plug-in to obtain surface radiation distribution data;

[0066] S24, based on the snow distribution and wind environment data and surface radiation distribution data of the ice shell surface, update the morphological construction information, load information and material performance information of the ice shell building and calculate the sublimation rate;

[0067] S25, based on the morphological construction information, load information and material performance information of the ice shell building and the sublimation rate, calculate the strain energy in the early and late service stages through the Karamba3D plug-in, and then calculate the weather resistance index of the ice shell building;

[0068] S26, taking the initial strain energy and the weather resistance index as the target, realize parameter optimization through the NSGA-II algorithm of the Wallacei plug-in.

[0069] Specifically, the purpose of establishing an agent model in S1 is to get rid of the influence of the designer's subjective intention on the usability of the model, and to realize high efficiency and universality. Therefore, this step does not directly output the prediction results of the optimization target through the building design parameters. In terms of features, it can be extracted from the building space form. Although the building form is generated by design parameters, its own features are universal. In terms of labels, wind environment and snow distribution are selected. These two are generally simulated by CFD, which is the most time-consuming step in performance index acquisition, and is also a necessary microclimate environmental factor in weather resistance mechanism. Therefore, the purpose of establishing an agent model in this step is to predict the surface wind environment and snow distribution through the spatial morphological features of the shell.

[0070] In S25, the load information is determined by the snow distribution, and the snow-covered part is set according to the snow depth of 0.3m and the density of 450kg / m 3 The main variable parameter of material performance is the elastic modulus, and the elastic modulus in the late service period is set to 70% of the initial period. The initial material parameter settings are shown in Table 1:

[0071] Table 1

[0072]

[0073] Through Figure 1 and Figure 2It can be known that the application optimizes the shape configuration and material performance simultaneously through the agent model and simulation analysis, so as to achieve the comprehensive goals of strain energy minimization (structural stability) and weather resistance improvement (environmental adaptability).

[0074] Figure 3 The results of the S26 parameter optimization are shown, and by comparing the trade-off relationship between the strain energy and weather resistance index at the initial service stage under different design parameters, the optimal balance point of the shell shape between the structural performance and weather resistance can be obtained.

[0075] In S11, in the shape feature, the Z-axis slope is obtained by:

[0076]

[0077] obtained, wherein S Z is the Z-axis slope, N is the number of neighborhood unit surfaces, is the unit normal vector, is the unit vector in the Z-axis direction;

[0078] The normalized rib distance is obtained by:

[0079]

[0080] obtained, wherein is the normalized rib distance, D j,i is the distance from the point j to the rib i, D j is all the shortest distance data;

[0081] The normalized height is obtained by:

[0082]

[0083] obtained, wherein H z is the point z coordinate data, H j is the z coordinate data of all points;

[0084] The maximum principal curvature and the minimum principal curvature are obtained by:

[0085]

[0086] obtained, wherein C(v) is the principal curvature tensor, |B| is the area of the vertex 1-ring neighborhood, beta (e) is the included angle of the surface normal vectors on both sides of the edge e, |e intersection B| is the length of the edge e in the neighborhood B, is the tensor product in the edge direction.

[0087] Specifically, the Z-axis slope is the average of the angle between the unit normal vector of the point-ring neighborhood unit surface and the z-axis, and is normalized to the slope range of 0° (horizontal plane) to 90° (vertical plane); the normalized rib distance is Min-Max normalized data of the shortest distance from the vertex to all rib lines; and the normalized height is Min-Max normalized data of the z coordinate of the vertex.

[0088] In S11, the Y-axis slope in the spatial feature is obtained by:

[0089]

[0090] obtained, wherein S y is the Y-axis slope, is the unit vector in the y-axis direction;

[0091] The normalized wind direction distance is obtained by:

[0092]

[0093] obtained, wherein is the normalized wind direction distance, D w is the vertical distance data from the point to the wind inflow boundary surface, D n is the distance data of all points.

[0094] Specifically, the Y-axis slope is the average of the angle between the unit normal vector of the point-ring neighborhood unit surface and the y-axis, and is normalized to the slope range of 0° (same direction surface) to 180° (opposite direction surface); and the normalized wind direction distance is Min-Max normalized of the distance from the point to the wind inflow boundary surface:

[0095] In S13, the artificial neural network proxy model is specifically constructed as follows: for snow distribution prediction, a 4-factor model containing the Z-axis slope, the normalized height, the distance from the rib line, and the minimum principal curvature is selected; and for wind environment prediction, a 7-factor model containing the Z-axis slope, the Y-axis slope, the normalized height, the distance from the rib line, the distance from the wind direction, the maximum principal curvature, and the minimum principal curvature is selected.

[0096] Specifically, the training of the model is implemented using the Tensorflow 2.8 framework of the Python 3.9 platform. Taking the airbag template ice shell building as the object, 5F models (Z-axis slope, normalized height, distance from rib line, maximum principal curvature, and minimum principal curvature), 4F1 models (Z-axis slope, distance from rib line, maximum principal curvature, and minimum principal curvature), 4F2 models (Z-axis slope, normalized height, distance from rib line, and minimum principal curvature), and 3F models (Z-axis slope, distance from rib line, and minimum principal curvature) are set for the snow distribution label. The Loss and Accuracy in the training process are compared as follows: Figure 4as shown; by Figure 5 The typical combination of shuttle-shaped shell morphology as shown is verified, and the 4F2 model is selected as the best prediction effect.

[0097] Taking the airbag template ice shell building as the object, 8F model (Z-axis slope, X-axis slope, Y-axis slope, normalized height, distance from rib line, distance from wind direction, maximum principal curvature, minimum principal curvature), 7F1 model (Z-axis slope, Y-axis slope, normalized height, distance from rib line, distance from wind direction, maximum principal curvature, minimum principal curvature), 7F2 model (Z-axis slope, X-axis slope, Y-axis slope, normalized height, distance from rib line, distance from wind direction, minimum principal curvature), 6F model (Z-axis slope, Y-axis slope, normalized height, distance from rib line, distance from wind direction, minimum principal curvature), 5F1 model (Z-axis slope, Y-axis slope, distance from rib line, distance from wind direction), 5F2 model (Z-axis slope, Y-axis slope, distance from rib line, distance from wind direction, minimum principal curvature), 4F model (Z-axis slope, Y-axis slope, distance from rib line, distance from wind direction) are set for the wind environment label, and the Loss, R 2 and MAE are compared during the training process. Figure 6 Figure 7 The typical combination of shuttle-shaped shell morphology as shown is verified, and the 7F1 model is selected as the best prediction effect.

[0098] In S21, the weather resistance index of the ice shell building is obtained by:

[0099] W=(U E -U S ) / U S

[0100] , where W is the weather resistance index of the ice shell building, U E is the initial stage elastic energy, and U S is the elastic energy at a specified service time.

[0101] The elastic energy is obtained by:

[0102]

[0103] , where F i is the load of the node, and μ i is the displacement of the node.

[0104] Further, in S24, the sublimation rate is obtained by:

[0105]

[0106] T s =T a +0.0559q G ​

[0107] obtained, where E FRI is the sublimation rate, v is the wind speed, ω is the paper fiber concentration, H is the cumulative sublimation height, T s is the material surface temperature, T a is the air temperature, q G is the solar radiation intensity.

Claims

1. A method for optimizing the weather resistance of a balloon template ice shell building based on a proxy model, characterized in that, The method comprises the following steps: S1, establishing a proxy model of spatial form features and microclimate response; S11, constructing an ice shell building case set through orthogonal experiment design, and extracting form features and spatial features; The form features include Z-axis slope, normalized rib line distance, normalized height, maximum principal curvature and minimum principal curvature; The spatial features include Y-axis slope and normalized wind direction distance; S12, performing CFD simulation of wind environment and snow distribution on the ice shell building case set to obtain label information; S13, model training based on the form features, spatial features and label information to construct an artificial neural network proxy model, The artificial neural network comprises 1 input layer, 4 hidden layers and 1 output layer; S2, calculating weather resistance index and multi-objective optimization; S21, defining the weather resistance index of the ice shell building; S22, constructing an ice shell building parameterized model using the Grasshopper platform to extract the form features and spatial features; S23, calling back the proxy model in S1 through the gh_Cpython plug-in to input the form features and spatial features, and obtaining snow distribution and wind environment data on the ice shell surface; combining the Ladybug plug-in to obtain surface radiation distribution data; S24, updating the form construction information, load information and material performance information of the ice shell building based on the snow distribution and wind environment data and surface radiation distribution data on the ice shell surface, and calculating the sublimation rate; S25, based on the form construction information, load information and material performance information of the ice shell building and the sublimation rate, calculating the strain energy at the initial and later stages of service through the Karamba3D plug-in, and then calculating the weather resistance index of the ice shell building; S26, taking the initial strain energy and the weather resistance index as the target, and realizing parameter optimization through the NSGA-II algorithm of the Wallacei plug-in; In S21, the weather resistance index of the ice shell building is calculated by: The elastic energy is calculated by: In S24, the sublimation rate is calculated by: The normalized rib line distance is calculated by: ； obtained, wherein is an index of weather resistance for ice rink buildings, is an elastic energy at an initial stage, is an elastic energy at a certain specified service time; The normalized height is calculated by: ； obtained, wherein is a load of the node, is a displacement of the node; The maximum principal curvature and minimum principal curvature are calculated by: ； ； ； obtained, wherein is the sublimation rate, is the wind speed, is the paper fiber concentration, is the cumulative sublimation height, is the material surface temperature, is the air temperature, is the solar radiation intensity.

2. The method according to claim 1, wherein, In S11, in the shape feature, the Z Axial slope is passed through: ； obtained, wherein, is axis slope, is the number of neighboring cell faces, is the unit normal vector, is Z unit vector in the axis direction; In S11, in the spatial features, ； obtained, wherein, is the normalized rib distance, is the point distance to the rib line, is all shortest distance data; The normalized wind direction distance is calculated by: ； obtained, wherein z coordinate data for the point, z coordinate data for all points; 4. The weather resistance optimization design method for the airbag template ice shell building based on the proxy model according to claim 1, ； obtained, where, is the principal curvature tensor, is the area of the vertex 1-ring neighborhood, is the edge the included angle between the normal vectors of the two sides, edge in the neighborhood the length within, is the tensor product of the edge direction.

3. The method according to claim 1, wherein, In S13, the construction of the artificial neural network proxy model is specifically: for snow distribution prediction, a 4-factor model is selected, which contains the Z-axis slope, the normalized height, the distance from the rib line and the minimum principal curvature; for wind environment prediction, a 7-factor model is selected, which contains the Z-axis slope, the Y-axis slope, the normalized height, the distance from the rib line, the distance from the wind direction, the maximum principal curvature and the minimum principal curvature. ； obtained, wherein is the Y-axis slope, is the unit vector in the y-axis direction; ​ ； obtained, wherein is the normalized wind direction distance, is the vertical distance data of the point to the wind inflow interface, is the distance data of all points. ​ characterized in that ​

Images