Method and device for designing building structural component dimensions based on heterogeneous graph neural network

Abstract

The present invention provides a method and device for designing the dimensions of building structure components based on a heterogeneous graph neural network, wherein the method comprises: obtaining a building structure design drawing to be processed, and converting the building structure design drawing into a heterogeneous graph; based on a pre-trained heterogeneous graph neural network model, predicting the component dimension data corresponding to each heterogeneous node according to the heterogeneous graph, and obtaining the building structure component dimension design result; wherein the heterogeneous graph neural network model is obtained by training and optimization based on a training sample data set composed of building structure design drawing samples and their corresponding actual component dimension data. The method represents the building structure components and their connection relationships through a heterogeneous graph, and uses the heterogeneous graph neural network model to extract features from the heterogeneous graph, which can not only quickly generate the dimension design of the building structure components and improve the design efficiency, but also ensure the precision of the dimension design and improve the design effect.

Description

Technical Field

[0001] The present invention relates to the field of intersectional technology of building structure design and artificial intelligence, and in particular to a method and device for designing building structure component dimensions based on a heterogeneous graph neural network. Background Art

[0002] At present, the process of building structure design mainly relies on the professional knowledge of engineers to preliminarily determine the size of components, and then make detailed adjustments through multiple model construction and calculations. This process not only has a high technical threshold, but also takes a long time, which is not conducive to the realization of automation, thus reducing the efficiency of structural design.

[0003] The use of computer-aided optimization methods can simplify manual operations and shorten the design process to a certain extent, but these methods usually require a long optimization time and mainly focus on the overall optimization goals, with less consideration of the specific dimensional details of the components.

[0004] Therefore, it is necessary to solve the problems of low efficiency and low design refinement of existing building structure size design methods. Summary of the invention

[0005] The present invention provides a method and device for designing the dimensions of building structure components based on a heterogeneous graph neural network, so as to overcome the defects of the existing building structure dimension design methods, such as low working efficiency and low degree of refinement of design effect, realize the rapid design of the dimensions of building structure components, improve work efficiency, and ensure the refinement of dimension design.

[0006] On the one hand, the present invention provides a method for designing the size of building structure components based on a heterogeneous graph neural network, comprising: obtaining a building structure design drawing to be processed, and converting the building structure design drawing into a heterogeneous graph; wherein the heterogeneous graph includes heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent building structure components, and the heterogeneous edges represent the connection relationship between the building structure components; based on a pre-trained heterogeneous graph neural network model, predicting the component size data corresponding to each heterogeneous node according to the heterogeneous graph, and obtaining a building structure component size design result; wherein the heterogeneous graph neural network model is obtained by training and optimizing a training sample data set composed of building structure design drawing samples and their corresponding actual component size data.

[0007] Furthermore, the step of converting the building structure design drawings into a heterogeneous graph specifically includes: extracting the building structure component axes corresponding to the building structure components in the building structure design drawings; determining multiple types of heterogeneous nodes and heterogeneous edges in the heterogeneous graph according to the building structure component axes and their positional relationships to form an initial heterogeneous graph; determining structural component features corresponding to different building structure components in the building structure design drawings, and storing the structural component features as node attributes in the heterogeneous nodes of the initial heterogeneous graph to obtain a final heterogeneous graph.

[0008] Furthermore, the structural component characteristics include structural characteristics and component characteristics; wherein the structural characteristics include the seismic design conditions, height information, size information and shape information of the entire building structure, and the component characteristics include the absolute position, relative position, material properties, load conditions and known dimensions of the building structure components.

[0009] Furthermore, the pre-trained heterogeneous graph-based neural network model predicts component size data corresponding to each heterogeneous node according to the heterogeneous graph, including: performing message transmission and information aggregation on the heterogeneous nodes corresponding to each type of heterogeneous edges in the heterogeneous graph, and updating the initial heterogeneous node features; aggregating the initial heterogeneous node features of heterogeneous nodes corresponding to different types of heterogeneous edges to obtain target heterogeneous node features; and converting the target heterogeneous node features to obtain component size data corresponding to each heterogeneous node.

[0010] Furthermore, training and optimizing the heterogeneous graph neural network model specifically includes: constructing a training sample data set, and converting the building structure design drawing samples therein into a heterogeneous graph representation; using the heterogeneous graph representation as model input, using the predicted dimension data as model output, and using the difference between the predicted dimension data and the actual component dimension data as training loss, iteratively optimizing the heterogeneous graph neural network model through a preset loss function, and obtaining a pre-trained heterogeneous graph neural network model.

[0011] Furthermore, the obtaining of the building structure component size design result then includes: adjusting the building structure component size design result according to the module standard, floor relationship and structural specification of the building structure component to obtain the target design size of the building structure component.

[0012] Furthermore, the method of obtaining a target design size of a building structure component includes: feeding back the target design size of the building structure component to the heterogeneous graph to obtain a target heterogeneous graph; redrawing a building structure design drawing according to the target heterogeneous graph to obtain a target building structure design drawing; constructing a corresponding building structure model according to the target building structure design drawing; and analyzing the mechanical properties and material usage of the building structure based on the building structure model to verify the design result of the building structure component size.

[0013] In a second aspect, the present invention also provides a device for designing the size of building structure components based on a heterogeneous graph neural network, comprising: a heterogeneous graph conversion module, used to obtain the building structure design drawings to be processed, and convert the building structure design drawings into a heterogeneous graph; wherein the heterogeneous graph includes heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent building structure components, and the heterogeneous edges represent the connection relationship between the building structure components; a component size prediction module, used to predict the component size data corresponding to each heterogeneous node according to the heterogeneous graph based on a pre-trained heterogeneous graph neural network model, and obtain the building structure component size design result; wherein the heterogeneous graph neural network model is trained and optimized based on a training sample data set composed of building structure design drawing samples and their corresponding actual component size data.

[0014] In a third aspect, the present invention also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for designing the dimensions of building structural components based on a heterogeneous graph neural network as described in any one of the above is implemented.

[0015] In a fourth aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a method for designing the dimensions of building structural components based on a heterogeneous graph neural network as described in any one of the above.

[0016] The method for designing the size of building structure components based on heterogeneous graph neural network provided by the present invention obtains the building structure design drawings to be processed and converts the building structure design drawings into heterogeneous graphs; wherein the heterogeneous graphs include heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent the building structure components, and the heterogeneous edges represent the connection relationship between the building structure components; and based on the pre-trained heterogeneous graph neural network model, the component size data corresponding to each heterogeneous node is predicted according to the heterogeneous graph to obtain the building structure component size design result; wherein the heterogeneous graph neural network model is trained and optimized based on the training sample data set composed of the building structure design drawing samples and the corresponding actual component size data. The method represents the building structure components and their connection relationship through heterogeneous graphs, and uses the heterogeneous graph neural network model to extract features from the heterogeneous graphs, which can not only quickly generate the size design of the building structure components and improve the design efficiency, but also ensure the precision of the size design and improve the design effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 It is a flow chart of a method for designing dimensions of building structural components based on a heterogeneous graph neural network provided in an embodiment of the present invention.

[0019] Figure 2 It is a schematic diagram of conversion of a heterogeneous graph provided by an embodiment of the present invention.

[0020] Figure 3 It is a schematic diagram of the model framework of the heterogeneous graph neural network model provided by an embodiment of the present invention.

[0021] Figure 4 It is a schematic diagram of the overall training optimization of the heterogeneous graph neural network model provided by an embodiment of the present invention.

[0022] Figure 5 It is a schematic diagram of floor assembly of a building structure provided by an embodiment of the present invention.

[0023] Figure 6 It is a schematic diagram of the overall process of the method for designing the size of building structure components based on the heterogeneous graph neural network model provided by an embodiment of the present invention.

[0024] Figure 7 It is a structural schematic diagram of a building structure component size design device based on a heterogeneous graph neural network provided in an embodiment of the present invention.

[0025] Figure 8 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] It should be noted that the current building structure design process mainly relies on the professional knowledge of engineers to preliminarily determine the size of components, and then make detailed adjustments through multiple model construction and calculations. This process not only has a high technical threshold, but also takes a long time, which is not conducive to the realization of automation, thus reducing the efficiency of building structure design.

[0028] The use of computer-aided optimization methods can simplify manual operations and shorten the design process to a certain extent, but these methods usually require a long optimization time and mainly focus on the overall optimization goals, with less consideration of the specific dimensional details of the components.

[0029] In view of this, an embodiment of the present invention proposes a method for designing the size of building structure components based on a heterogeneous graph neural network. Specifically, Figure 1 A schematic flow chart of a method for designing dimensions of building structural components based on a heterogeneous graph neural network provided in an embodiment of the present invention is shown.

[0030] like Figure 1 As shown, the method includes steps S110-S120, and these steps will be described in detail below.

[0031] S110, obtaining a building structure design drawing to be processed, and converting the building structure design drawing into a heterogeneous graph; wherein the heterogeneous graph includes heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent building structure components, and the heterogeneous edges represent connection relationships between building structure components.

[0032] It is understandable that, for the architectural structure design drawings to be dimensionally designed, they need to be converted into the form of heterogeneous graphs.

[0033] Specifically, it is first necessary to extract the axes of the building structure components in the building structure design drawings, and the axes of the building structure components correspond to the building structure components.

[0034] In a heterogeneous graph, there are heterogeneous nodes and edges. Heterogeneous nodes represent building structural components, while heterogeneous edges represent the connection relationship between building structural components. Building structural components and the connection relationship between building structural components include many different types, among which the types of building structural components include beam components, plate components, shear wall components, and column components, and the types of connection relationships include beam-beam connection relationship, beam-column connection relationship, and beam-plate connection relationship.

[0035] After extracting the axis of the building structure components, the corresponding building structure components and the connection relationship between the building structure components can be obtained, and then the heterogeneous nodes and heterogeneous edges of the heterogeneous graph can be determined to obtain an initial heterogeneous graph.

[0036] On this basis, it is worth mentioning that different building structural components have different structural component characteristics, such as the location, material properties, load conditions, etc. of the building structural components. This embodiment pre-acquires different structural component characteristics corresponding to different building structural components, and stores these structural component characteristics as heterogeneous node attributes in the heterogeneous nodes of the initial heterogeneous graph, thereby obtaining the final heterogeneous graph.

[0037] S120, based on a pre-trained heterogeneous graph neural network model, predict the component size data corresponding to each heterogeneous node according to the heterogeneous graph to obtain the building structure component size design result; wherein the heterogeneous graph neural network model is trained and optimized based on a training sample data set consisting of building structure design drawing samples and their corresponding actual component size data.

[0038] It can be understood that, based on obtaining the building structure design drawings to be processed in step S110 and converting the building structure design drawings into heterogeneous graphs, step S120 further inputs the heterogeneous graph into a pre-trained heterogeneous graph neural network model, and through feature extraction, the component size data corresponding to each heterogeneous node of the output can be obtained, that is, the building structure component size design result.

[0039] Among them, the heterogeneous graph neural network is trained and optimized based on a training sample data set composed of samples of building structure design drawings and the actual component size data corresponding to the samples.

[0040] When training a heterogeneous graph neural network, it is necessary to convert the building structure design drawings into corresponding heterogeneous graph representations in accordance with step S110 to serve as model input during training.

[0041] It is worth mentioning that this embodiment uses a heterogeneous graph neural network model to predict the component size data corresponding to each heterogeneous node, which can better capture the complex structure of the data, more comprehensively describe the complex relationship between the data, and more clearly display the hierarchical relationship and association relationship between the data. It has strong generalization ability, expression ability and interpretability.

[0042] It should also be noted that in order to make the predicted building structure component size design results more in line with the structural design specifications, the building structure component size design results can be further adjusted, which is not specifically limited here.

[0043] In this embodiment, by obtaining the architectural structure design drawings to be processed, and converting the architectural structure design drawings into a heterogeneous graph; wherein the heterogeneous graph includes heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent the architectural structure components, and the heterogeneous edges represent the connection relationship between the architectural structure components; and based on the pre-trained heterogeneous graph neural network model, the component size data corresponding to each heterogeneous node is predicted according to the heterogeneous graph to obtain the architectural structure component size design result; wherein the heterogeneous graph neural network model is trained and optimized based on the training sample data set composed of the architectural structure design drawing samples and the corresponding actual component size data. This method uses a heterogeneous graph to represent the architectural structure components and their connection relationship, and uses the heterogeneous graph neural network model to extract features from the heterogeneous graph, which can not only quickly generate the size design of the architectural structure components and improve the design efficiency, but also ensure the precision of the size design and improve the design effect.

[0044] On the basis of the above-mentioned embodiment, the process of converting the architectural structure design drawings into heterogeneous graphs will be described in detail below.

[0045] The steps of converting a building structure design drawing into a heterogeneous graph specifically include: extracting the building structure component axes corresponding to the building structure components in the building structure design drawings; determining multiple types of heterogeneous nodes and heterogeneous edges in the heterogeneous graph according to the building structure component axes and their positional relationships to form an initial heterogeneous graph; determining structural component features corresponding to different building structure components in the building structure design drawings, and storing the structural component features as node attributes in the heterogeneous nodes of the initial heterogeneous graph to obtain a final heterogeneous graph.

[0046] It is understandable that in order to convert the architectural structure design drawings into the form of a heterogeneous graph, firstly, the architectural structure components and their corresponding architectural structure component axes in the architectural structure design drawings are extracted through automated data extraction operations to determine multiple types of heterogeneous nodes in the heterogeneous graph. On this basis, multiple types of heterogeneous edges in the heterogeneous graph are determined according to the spatial position relationship of the axes of the architectural structure components, thereby obtaining an initial heterogeneous graph.

[0047] Among them, multiple types of heterogeneous nodes refer to different types of building structural components that should be represented by different types of heterogeneous nodes. Different types of building structural components include but are not limited to beam components, plate components, shear wall components, column components, and support components.

[0048] Multiple types of heterogeneous edges refer to different types of connection relationships between different types of building structural components, which should be represented by different types of heterogeneous edges. Taking the concrete frame structure as an example, according to the relative positions of the building structural components in space, the connection relationships between beam components, column components, and plate components can be classified into four basic types.

[0049] (1) When two beam components are connected to the same column component, a beam-beam connection relationship is formed between the two beam components.

[0050] (2) The connection relationship between beam components and column components is represented by the beam-column connection relationship.

[0051] (3) The connection relationship between the plate component and the beam components surrounding it is defined by the beam-plate connection relationship.

[0052] (4) The connection relationship between the plate component and the column components surrounding it is defined by the column-plate connection relationship.

[0053] On the basis of the initial heterogeneous graph, different node attributes can be assigned to the heterogeneous nodes in the initial heterogeneous graph, and the inherent attributes of the building structure components can be reflected through the node attributes. Specifically, the structural component features corresponding to different building structure components in the building structure design drawings are determined, and the structural component features are stored as node attributes in the initial heterogeneous graph to obtain the final heterogeneous graph.

[0054] Among them, structural component characteristics include structural characteristics and component characteristics. Structural characteristics include the overall seismic design conditions, height information, size information and shape information of the building structure, and component characteristics include the absolute position, relative position, material properties, load conditions and known dimensions of the building structure components.

[0055] Still taking the concrete frame structure as an example, for beam components and column components, their corresponding node attributes include seismic design conditions, absolute position, relative position, material grade, length and direction; for plate components, their corresponding node attributes include seismic design conditions, absolute position, relative position, material grade, area, load, room type and number of sides.

[0056] The absolute position in the component characteristics refers to the coordinate position of the midpoint of the component on the plane after the standard layer is centered. The relative position refers to the relative coordinate position of the midpoint of the component after normalization, as well as the total length of the standard layer in the horizontal and vertical directions. The height characteristics (height information) cover the starting floor of the current standard layer, the total number of floors and floor heights of the building. The earthquake characteristics (seismic design conditions) cover the peak ground acceleration of the design benchmark earthquake, site category, seismic grade and design earthquake grouping. These characteristics are important bases for the seismic design of concrete frame structures according to the Chinese Building Seismic Design Code.

[0057] Based on the above, the architectural structure design drawings can be converted into heterogeneous graphs. For details, please refer to Figure 2 , Figure 2 A schematic diagram of conversion of a heterogeneous graph provided by an embodiment of the present invention is shown.

[0058] In this embodiment, by extracting the axis of the building structure component corresponding to the building structure component in the building structure design drawing, and determining multiple types of heterogeneous nodes and heterogeneous edges in the heterogeneous graph according to the axis of the building structure component and its positional relationship, an initial heterogeneous graph is formed, and then the structural component features corresponding to different building structure components in the building structure design drawing are determined, and the structural component features are stored as node attributes in the heterogeneous nodes of the initial heterogeneous graph to obtain the final heterogeneous graph, thereby, based on the pre-trained heterogeneous graph neural network model, the component size data corresponding to each heterogeneous node is predicted according to the heterogeneous graph to obtain the building structure component size design result. This method uses a heterogeneous graph to represent the building structure components and their connection relationships, and uses the heterogeneous graph neural network model to extract features from the heterogeneous graph, which can not only quickly generate the size design of the building structure components and improve the design efficiency, but also ensure the precision of the size design and improve the design effect.

[0059] On the basis of the above-mentioned embodiments, the processing process of the heterogeneous graph by the heterogeneous graph neural network model will be described in detail below.

[0060] Based on the pre-trained heterogeneous graph neural network model, the component size data corresponding to each heterogeneous node is predicted according to the heterogeneous graph, including: performing message transmission and information aggregation on the heterogeneous nodes corresponding to each type of heterogeneous edge in the heterogeneous graph, and updating the initial heterogeneous node features; aggregating the initial heterogeneous node features of the heterogeneous nodes corresponding to different types of heterogeneous edges to obtain the target heterogeneous node features; converting the target heterogeneous node features to obtain the component size data corresponding to each heterogeneous node.

[0061] It can be understood that the process of processing heterogeneous graphs using heterogeneous graph neural network models is mainly divided into three stages.

[0062] (1) In the first stage, message transmission and information aggregation are performed on heterogeneous nodes corresponding to the same type of heterogeneous edges (such as beam-column connection, beam-plate connection, or beam-beam connection) to update the heterogeneous node features corresponding to each heterogeneous node and obtain the initial heterogeneous node features.

[0063] (2) In the second stage, feature aggregation is performed on heterogeneous nodes corresponding to different types of heterogeneous edges (e.g., beam-column connection, beam-slab connection, and beam-beam connection). That is, the initial heterogeneous node features of heterogeneous nodes corresponding to different types of heterogeneous edges are aggregated to obtain target heterogeneous node features.

[0064] (3) In the third stage, the multilayer perceptron network is used to transform the target heterogeneous node features to obtain the component size data corresponding to each heterogeneous node, that is, the component size data corresponding to each building structure component.

[0065] Specifically, in order to ensure that each heterogeneous node can effectively transmit and receive information, each heterogeneous node is given a self-connection structure.

[0066] Next, for different types of heterogeneous edges, graph neural network technology is used to achieve information exchange and aggregation. The graph neural network technology here covers heterogeneous graph neural network models, including but not limited to GCN (Graph Convolutional Network), GAT (Graph Attention Networks) and GraphSAGE (Graph Sample and Aggregation) and other methods.

[0067] Subsequently, for each heterogeneous node type, the initial heterogeneous node features aggregated from multiple heterogeneous edges are summed and further transformed through a multi-layer perceptron network. The depth and learning ability of the model can also be increased through residual connection technology.

[0068] After multiple rounds of iterations, the process of processing the heterogeneous graph by the heterogeneous graph neural network model can be completed, and the preliminary prediction of the size of the building structure components can be achieved.

[0069] Figure 3 FIG. 1 shows a schematic diagram of a model framework of a heterogeneous graph neural network model provided by an embodiment of the present invention. Figure 3 As shown, the heterogeneous graph neural network model mainly includes graph neural network and multi-layer perceptron. The graph neural network includes but is not limited to the GCN, GAT and GraphSAGE mentioned above.

[0070] Among them, graph neural networks are used to realize the exchange and convergence of information in heterogeneous graphs, and multi-layer perceptrons are used to realize the conversion of target heterogeneous node features to component size data.

[0071] In this embodiment, the initial heterogeneous node features are updated by performing message transmission and information aggregation on the heterogeneous nodes corresponding to each type of heterogeneous edge in the heterogeneous graph, and the initial heterogeneous node features corresponding to the heterogeneous nodes of different types of heterogeneous edges are aggregated to obtain the target heterogeneous node features, and then the target heterogeneous node features are converted to obtain the component size data corresponding to each heterogeneous node. This method uses a heterogeneous graph to represent the building structure components and their connection relationships, and uses a heterogeneous graph neural network model to extract features from the heterogeneous graph, which can not only quickly generate the size design of the building structure components and improve the design efficiency, but also ensure the precision of the size design and improve the design effect.

[0072] On the basis of the above embodiments, the training optimization process of the heterogeneous graph neural network model will be described in detail below.

[0073] Training and optimizing a heterogeneous graph neural network model specifically includes: constructing a training sample data set, and converting the architectural structure design drawing samples therein into a heterogeneous graph representation; using the heterogeneous graph representation as the model input, using the predicted dimension data as the model output, and using the difference between the predicted dimension data and the actual component dimension data as the training loss, and iteratively optimizing the heterogeneous graph neural network model through a preset loss function to obtain a pre-trained heterogeneous graph neural network model.

[0074] It is understandable that in order to train and optimize the heterogeneous graph neural network model, we first collect historical building structure design drawings (i.e., building structure design drawing samples) and their corresponding actual component size data.

[0075] Then, according to the method of constructing a heterogeneous graph, the corresponding heterogeneous graph representation is obtained based on the conversion of the building structure design drawing sample, which will not be repeated here.

[0076] Next, the actual component size data and the corresponding heterogeneous graph representation are used to train and optimize the heterogeneous graph neural network model. Specifically, in each training process, the heterogeneous graph representation is used as the model input, the predicted size data is used as the model output, and the difference between the predicted size data and the actual component size data is used as the training loss. The heterogeneous graph neural network model is iteratively optimized through a preset loss function to obtain a pre-trained heterogeneous graph neural network model.

[0077] Among them, the preset loss function can be a mean square error loss function or other loss functions, which are not specifically limited here.

[0078] Furthermore, in a specific embodiment, Figure 4A schematic diagram of overall training optimization of the heterogeneous graph neural network model provided by an embodiment of the present invention is shown.

[0079] Based on actual engineering projects, this embodiment selects 97 concrete frame buildings, including 308 standard floors, to form a training sample data set. Considering that in a concrete frame structure, the secondary beam has a relatively small impact on the overall structural performance, and the size design of the secondary beam is mainly related to the load of the slab, the secondary beam is not taken into consideration, but the design of the main beam is focused on.

[0080] Afterwards, the CAD drawings are converted into heterogeneous graph representations using heterogeneous graph representations. Different types of heterogeneous nodes are used to represent beam components, column components, and plate components, respectively, and the structural component features corresponding to each type of component are stored in the corresponding heterogeneous nodes. The heterogeneous graph neural network model is used to learn the design pattern of the size of concrete frame structure components.

[0081] The training process uses the mean square error loss function to measure the deviation between the model prediction value and the actual observation value, which is used as the basis for optimizing the model performance. Regression analysis is used for prediction to ensure the continuity and orderliness of the prediction results.

[0082] In this embodiment, by constructing a training sample data set, and converting the architectural structure design drawing samples therein into a heterogeneous graph representation, then taking the heterogeneous graph representation as the model input, taking the predicted size data as the model output, and taking the difference between the predicted size data and the actual component size data as the training loss, the heterogeneous graph neural network model is iteratively optimized through a preset loss function to obtain a pre-trained heterogeneous graph neural network model, and then based on the pre-trained heterogeneous graph neural network model, the component size data corresponding to each heterogeneous node is predicted according to the heterogeneous graph to obtain the building structure component size design result. This method uses a heterogeneous graph to represent the building structure components and their connection relationships, and uses the heterogeneous graph neural network model to extract features from the heterogeneous graph, which can not only quickly generate the size design of the building structure components and improve the design efficiency, but also ensure the precision of the size design and improve the design effect.

[0083] On the basis of the above-mentioned embodiments, further, the post-processing of the design results of the building structure component dimensions will be described in detail below.

[0084] The size design result of the building structure component is obtained, and then the size design result of the building structure component is adjusted according to the modulus standard, floor relationship and structural specification of the building structure component to obtain the target design size of the building structure component.

[0085] It is understandable that the size design result of building structural components predicted by the heterogeneous graph neural network model is a close to optimal design result, but minor adjustments are still needed to generate a size design solution suitable for actual engineering.

[0086] First, according to the structural modulus requirements, the component size data is rounded. Specifically, based on the conventional experience of engineering design and historical statistical data, the modulus standards of different building structural components are determined (for example, the modulus standard of beam components and column components is determined to be 50 mm, and the modulus standard of plate components is determined to be 20 mm), and then the size of the building structural components is adjusted to an integer multiple of the modulus standard.

[0087] Then, based on the relationship between the floors, the size of the column components is further adjusted to ensure that the size of the upper columns of the adjacent floors is smaller than the size of the lower columns, that is, to ensure that the cross-sectional size of the columns of the lower building is not smaller than the cross-sectional size of the columns of the upper building.

[0088] Finally, according to the requirements of the structural specifications, the sizes of components that do not meet the specifications are adjusted to meet the specifications. This means that when designing a building structure, the size of the building structure needs to meet certain conditions, such as earthquake-resistant design conditions and wind-resistant design conditions, so the size of the components that do not meet the corresponding conditions is adjusted.

[0089] After the above adjustments, the target design dimensions of the building structural components can be obtained.

[0090] Furthermore, after obtaining the target design size of the building structure component, the method then includes: feeding back the target design size of the building structure component into the heterogeneous graph to obtain the target heterogeneous graph; redrawing the building structure design drawings according to the target heterogeneous graph to obtain the target building structure design drawings; constructing the corresponding building structure model according to the target building structure design drawings; and analyzing the mechanical properties and material usage of the building structure based on the building structure model to verify the design results of the building structure component sizes.

[0091] Specifically, first, the target design dimensions of the building structure components are stored in a heterogeneous graph representing the structure to obtain a target heterogeneous graph; then, the target heterogeneous graph is used to redraw the corresponding building structure design drawings, namely, the target building structure design drawings.

[0092] Next, the target building structure design drawings are assembled through reference points to construct a building structure model, and the building structure model is used to analyze the mechanical properties and material usage of the current structure to achieve the effect of verifying the design results of the building structure component dimensions.

[0093] Taking concrete frame structure as an example, Figure 5 A schematic diagram of floor assembly of a building structure provided by an embodiment of the present invention is shown.

[0094] like Figure 5 As shown, after adjusting the design results of the building structure component dimensions, the graphic information of each standard layer needs to be converted into a CAD drawing format again in order to generate a detailed dimension layout drawing of the building structure component.

[0095] Then, by applying parametric modeling technology, multiple standard layers in the target building are mapped into a unified coordinate system through reference points.

[0096] Then, the parametric modeling software YJK-GAMA is used to integrate these standard layers, automatically configure the global parameters of the entire structure, the local parameters of each standard layer, and the load data, so as to construct the architectural structure model of the entire building.

[0097] Finally, based on the building structure model, the mechanical properties and material usage of the building structure are analyzed. This process is a test of the design results of the heterogeneous graph neural network model to ensure the safety and reliability of the resulting structure.

[0098] In this embodiment, after obtaining the building structure component size design results, a series of post-processing is performed on the building structure component size design results, thereby further ensuring the safety and reliability of the building structure under the predicted component size data.

[0099] in addition, Figure 6 The overall process diagram of the building structure component size design method based on the heterogeneous graph neural network model provided by an embodiment of the present invention is shown.

[0100] like Figure 6 As shown, first, the building structure component axes corresponding to the building structure components in the building structure design drawings to be processed are extracted, and the building structure design drawings are converted into the form of heterogeneous graphs, and the corresponding structural component features are stored in the heterogeneous nodes.

[0101] Then, the heterogeneous graph representing the structure is input into the trained heterogeneous graph neural network model to obtain the component size data corresponding to each heterogeneous node, that is, the design result of the building structure component size.

[0102] Furthermore, empirical rules are used to further adjust the design results of building structural components to make them more consistent with structural design specifications, and obtain the target design dimensions of building structural components.

[0103] Finally, the target design dimensions of the building structural components are fed back into the heterogeneous graph, and multiple floors of the same building are automatically assembled and modeled to verify the dimensional design results.

[0104] Corresponding to the method for designing the size of building structure components based on heterogeneous graph neural networks described in the above embodiments, the present invention also provides a device for designing the size of building structure components based on heterogeneous graph neural networks.

[0105] Specifically, Figure 7 A schematic structural diagram of a building structure component size design device based on a heterogeneous graph neural network provided in an embodiment of the present invention is shown.

[0106] like Figure 7 As shown, the device includes: a heterogeneous graph conversion module 710, which is used to obtain the building structure design drawings to be processed and convert the building structure design drawings into a heterogeneous graph; wherein the heterogeneous graph includes heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent building structure components, and the heterogeneous edges represent the connection relationship between the building structure components; a component size prediction module 720, which is used to predict the component size data corresponding to each heterogeneous node according to the heterogeneous graph based on a pre-trained heterogeneous graph neural network model, and obtain the building structure component size design result; wherein the heterogeneous graph neural network model is obtained by training and optimization based on a training sample data set composed of building structure design drawing samples and their corresponding actual component size data.

[0107] In this embodiment, the building structure design drawings to be processed are obtained through the heterogeneous graph conversion module 710, and the building structure design drawings are converted into a heterogeneous graph; wherein the heterogeneous graph includes heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent the building structure components, and the heterogeneous edges represent the connection relationship between the building structure components; the component size prediction module 720 predicts the component size data corresponding to each heterogeneous node according to the heterogeneous graph based on the pre-trained heterogeneous graph neural network model, and obtains the building structure component size design result; wherein the heterogeneous graph neural network model is trained and optimized based on the training sample data set composed of the building structure design drawing samples and the corresponding actual component size data. The device uses a heterogeneous graph to represent the building structure components and their connection relationship, and uses the heterogeneous graph neural network model to extract features from the heterogeneous graph, which can not only quickly generate the size design of the building structure components and improve the design efficiency, but also ensure the precision of the size design and improve the design effect.

[0108] It should be noted that the device for designing the size of building structure components based on a heterogeneous graph neural network provided in this embodiment can correspond to the method for designing the size of building structure components based on a heterogeneous graph neural network described in the above embodiments, and will not be repeated here.

[0109] Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8As shown, the electronic device may include: a processor 810, a communication interface 820, a memory 830 and a communication bus 840, wherein the processor 810, the communication interface 820 and the memory 830 communicate with each other through the communication bus 840. The processor 810 may call the logic instructions in the memory 830 to execute the building structure component size design method based on the heterogeneous graph neural network, the method comprising: obtaining the building structure design drawing to be processed, and converting the building structure design drawing into a heterogeneous graph; wherein the heterogeneous graph comprises heterogeneous nodes and heterogeneous edges, wherein the heterogeneous nodes represent the building structure components, and the heterogeneous edges represent the connection relationship between the building structure components; based on the pre-trained heterogeneous graph neural network model, the component size data corresponding to each heterogeneous node is predicted according to the heterogeneous graph, and the building structure component size design result is obtained; wherein the heterogeneous graph neural network model is obtained by training and optimizing the training sample data set composed of the building structure design drawing samples and the corresponding actual component size data.

[0110] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0111] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the building structure component size design method based on a heterogeneous graph neural network provided by the above-mentioned methods, the method comprising: obtaining a building structure design drawing to be processed, and converting the building structure design drawing into a heterogeneous graph; wherein the heterogeneous graph comprises heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent building structure components, and the heterogeneous edges represent the connection relationship between the building structure components; based on a pre-trained heterogeneous graph neural network model, the component size data corresponding to each heterogeneous node is predicted according to the heterogeneous graph to obtain the building structure component size design result; wherein the heterogeneous graph neural network model is obtained by training and optimization based on a training sample data set composed of building structure design drawing samples and their corresponding actual component size data.

[0112] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0113] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for designing the size of building structural components based on heterogeneous graph neural networks, characterized in that: include: Acquire a building structure design drawing to be processed, and convert the building structure design drawing into a heterogeneous graph; wherein the heterogeneous graph includes heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent building structure components, and the heterogeneous edges represent connection relationships between building structure components; Based on a pre-trained heterogeneous graph neural network model, the component size data corresponding to each heterogeneous node is predicted according to the heterogeneous graph to obtain the building structure component size design result; wherein the heterogeneous graph neural network model is trained and optimized based on a training sample data set consisting of building structure design drawing samples and their corresponding actual component size data.

2. The method for designing building structure components based on heterogeneous graph neural network according to claim 1 is characterized in that: The step of converting the architectural structure design drawing into a heterogeneous graph specifically includes: Extracting the axis of the building structure component corresponding to the building structure component in the building structure design drawing; According to the axes of the building structural components and their positional relationships, multiple types of heterogeneous nodes and heterogeneous edges in the heterogeneous graph are determined to form an initial heterogeneous graph; The structural component features corresponding to different building structural components in the building structure design drawings are determined, and the structural component features are stored as node attributes in the heterogeneous nodes of the initial heterogeneous graph to obtain a final heterogeneous graph.

3. The method for designing building structure components based on heterogeneous graph neural network according to claim 2 is characterized in that: The structural component features include structural features and component features; The structural features include the seismic design conditions, height information, size information and shape information of the building structure as a whole, and the component features include the absolute position, relative position, material properties, load conditions and known dimensions of the building structure components; The earthquake design conditions cover the peak ground acceleration of the design benchmark earthquake, site category, seismic resistance level and design earthquake grouping; the height information covers the starting floor of the current standard layer, the total number of floors and floor heights of the building; the absolute position is the coordinate position of the midpoint of the component on the plane after the standard layer is centered; the relative position is the relative coordinate position of the midpoint of the component after normalization, as well as the total length of the standard layer in the horizontal and vertical directions.

4. The method for designing building structure component dimensions based on heterogeneous graph neural network according to claim 1 is characterized in that: The pre-trained heterogeneous graph neural network model predicts the component size data corresponding to each heterogeneous node according to the heterogeneous graph, including: Perform message transmission and information aggregation on the heterogeneous nodes corresponding to each type of heterogeneous edge in the heterogeneous graph, and update the initial heterogeneous node features; Aggregate the initial heterogeneous node features of heterogeneous nodes corresponding to different types of heterogeneous edges to obtain the target heterogeneous node features; The target heterogeneous node features are converted to obtain component size data corresponding to each heterogeneous node.

5. The method for designing building structure component dimensions based on heterogeneous graph neural network according to claim 1 is characterized in that: Training and optimizing the heterogeneous graph neural network model specifically includes: Construct a training sample dataset and convert the architectural structure design drawing samples into heterogeneous graph representations; The heterogeneous graph representation is used as the model input, the predicted dimension data is used as the model output, and the difference between the predicted dimension data and the actual component dimension data is used as the training loss. The heterogeneous graph neural network model is iteratively optimized through a preset loss function to obtain a pre-trained heterogeneous graph neural network model.

6. The method for designing building structure components based on heterogeneous graph neural network according to any one of claims 1 to 5, characterized in that: The method of obtaining the design result of the building structure component size comprises: According to the modular standards, floor relationships and structural specifications of the building structural components, the size design results of the building structural components are adjusted to obtain the target design size of the building structural components.

7. The method for designing building structure components based on heterogeneous graph neural network according to claim 6 is characterized in that: The method of obtaining the target design size of the building structural component comprises: Feeding back the target design dimensions of the building structural components into the heterogeneous graph to obtain a target heterogeneous graph; Redraw the building structure design drawing according to the target heterogeneous graph to obtain the target building structure design drawing; According to the target building structure design drawings, a corresponding building structure model is constructed; Based on the building structure model, the mechanical properties and material usage of the building structure are analyzed to verify the design results of the size of the building structure components.

8. A device for designing the size of building structural components based on heterogeneous graph neural networks, characterized in that: include: A heterogeneous graph conversion module is used to obtain a building structure design drawing to be processed and convert the building structure design drawing into a heterogeneous graph; wherein the heterogeneous graph includes heterogeneous nodes and heterogeneous edges, the heterogeneous nodes represent building structure components, and the heterogeneous edges represent the connection relationship between the building structure components; The component size prediction module is used to predict the component size data corresponding to each heterogeneous node based on a pre-trained heterogeneous graph neural network model according to the heterogeneous graph, so as to obtain the building structure component size design result; wherein the heterogeneous graph neural network model is trained and optimized based on a training sample data set composed of building structure design drawing samples and their corresponding actual component size data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, it implements the method for designing the size of building structure components based on a heterogeneous graph neural network as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for designing the dimensions of building structural components based on a heterogeneous graph neural network as described in any one of claims 1 to 7 is implemented.

Images