Graph structure representation of BIM model and model error identification method and device

Abstract

The application relates to the technical field of BIM intelligent design and deep learning, in particular to a BIM model graph structure representation and model error identification method and device, wherein the method comprises the following steps: extracting semantic features, spatial features and topological features among various components according to component information of different types of components in a BIM model; generating corresponding graph structure representations based on the semantic features, the spatial features and the topological features, and constructing a training data set by using the graph structure representations; training a pre-trained graph neural network model by using the training data set to obtain a trained training model, and performing transfer learning by using the trained training model to identify an error identification target of the BIM model by using a transfer learning model. Thus, the problems that in the related art, the complex semantic, spatial and topological features of the BIM model bring difficult-to-ignore challenges, and the pre-training and transfer learning model cannot be perfectly suitable for the BIM model features are solved.

Description

Technical Field

[0001] This application relates to the fields of BIM (Building Information Modeling) intelligent design and deep learning technology, and in particular to a graphical representation of BIM models and a method and apparatus for model error identification. Background Technology

[0002] As a digital representation of building characteristics, BIM integrates a wealth of engineering data, including semantic features, spatial layout, material properties, and other relevant information. For example, BIM model data often contains a wealth of practical experience from engineers, such as standard ranges for door and window dimensions, and the spatial and connection relationships between various components. Therefore, extracting, learning, and reusing relevant design knowledge and patterns from this data is crucial for improving the level of building design, construction, and maintenance.

[0003] Among related technologies, machine learning can be used to train structured BIM model data to extract design features of the BIM model; deep learning can be used to learn and drive labeled data to extract corresponding features; pre-trained and transfer models can be used to perform self-supervised learning on unlabeled data to mine the inherent features and contextual information of the data, and then transfer a small amount of labeled data to a specific task.

[0004] However, despite significant progress in text and visual data processing, the complex semantic, spatial, and topological features unique to BIM models still present considerable challenges, making it impossible to perfectly apply pre-trained and transfer learning models that feature BIM models, thus requiring urgent improvement. Summary of the Invention

[0005] This application provides a graph structure representation of BIM models and a method and apparatus for model error identification, in order to solve the problems in related technologies, such as the challenges brought by the complex semantics, spatial and topological features unique to BIM models, which cannot be ignored, and the inability to be perfectly applied to the pre-training and transfer learning models of BIM model features.

[0006] The first aspect of this application provides a graph structure representation of a BIM model and a method for model error identification, applied in the model training stage. The method includes the following steps: extracting semantic features, spatial features, and topological features between different types of components based on component information in the BIM model; generating a graph structure representation of the BIM model based on the semantic features, spatial features, and topological features, and constructing a training dataset suitable for a pre-trained graph neural network model using the graph structure representation; training the pre-trained graph neural network model using the training dataset to obtain a trained model, and performing transfer learning on the trained model to obtain a transfer learning model, which is then used to identify error targets in the BIM model.

[0007] Optionally, in one embodiment of this application, the step of extracting semantic features, spatial features, and topological features between different types of components based on component information of different types of components in the BIM model includes: extracting the semantic features based on the component feature information of each component; extracting the spatial features based on the location information of each component and the local spatial relationship between adjacent components; and extracting the topological features based on the nesting relationship and connection relationship between different components, as well as the vertical relationship between the different components and the ground.

[0008] Optionally, in one embodiment of this application, generating a graph structure representation of the BIM model based on the semantic features, spatial features, and topological features includes: converting the different types of components into graph nodes in the graph structure representation, and obtaining the node attributes of the graph nodes based on the semantic features; obtaining graph edges in the graph structure representation based on the spatial features or the topological features, and obtaining the spatial edge attributes of the graph edges corresponding to the spatial features based on the local spatial relationships between adjacent components; obtaining the topological edge attributes of the graph edges corresponding to the topological features based on the nesting relationships, connection relationships, and vertical relationships in the topological features; and obtaining the graph structure representation based on at least one of the graph nodes, the node attributes, the graph edges, the spatial edge attributes, and the topological edge attributes.

[0009] Optionally, in one embodiment of this application, training the pre-trained graph neural network model using the training dataset to obtain a trained model includes: generating a corresponding occlusion dataset using the training dataset; inputting the occlusion dataset into the encoder of the pre-trained graph neural network model to generate a first high-dimensional representation of the graph nodes and / or a second high-dimensional representation of the edge attributes in the occlusion dataset using the encoder; inputting the first high-dimensional representation and / or the second high-dimensional representation into the decoder of the pre-trained graph neural network model to reconstruct the occluded graph nodes and occluded graph edges in the occlusion dataset using the decoder to obtain reconstructed graph nodes and reconstructed graph edges; and obtaining the trained model based on the training dataset, the occlusion dataset, the reconstructed graph nodes, and the reconstructed graph edges.

[0010] A second aspect of this application provides a graph structure representation of a BIM model and a method for identifying model errors, applied in the model transfer learning stage. The method includes the following steps: extracting semantic features, spatial features, and topological features between different types of components based on component information of different types of components in the target BIM model; generating a target graph structure representation of the target BIM model based on the semantic features, spatial features, and topological features; and inputting the target graph structure representation into a pre-trained transfer learning model to identify erroneous target identification results of the target BIM model, wherein the pre-trained transfer learning model is obtained through transfer learning from a trained training model.

[0011] Optionally, in one embodiment of this application, before inputting the target graph structure representation into the pre-trained transfer learning model, the method further includes: extracting semantic features, spatial features, and topological features between different types of components based on component information of different types of components in the BIM model containing the misidentified target; generating an error graph structure representation of the BIM model containing the misidentified target based on the semantic features, spatial features, and topological features, and constructing a transfer dataset suitable for the pre-trained transfer learning model using the error graph structure representation; and training the pre-trained transfer learning model using the transfer dataset until a preset transfer condition is met to obtain the trained transfer learning model.

[0012] A third aspect of this application provides a graph structure representation and error identification device for a BIM model, applied in the model training stage. The device includes: a first extraction module for extracting semantic features, spatial features, and topological features between different types of components based on component information in the BIM model; a first construction module for generating a graph structure representation of the BIM model based on the semantic features, spatial features, and topological features, and constructing a training dataset suitable for a pre-trained graph neural network model using the graph structure representation; and a first training module for training the pre-trained graph neural network model using the training dataset to obtain a trained model, and performing transfer learning on the trained model to obtain a transfer learning model, which is then used to identify error-identifying targets in the BIM model.

[0013] Optionally, in one embodiment of this application, the first extraction module includes: a first extraction unit, configured to extract the semantic features based on the component feature information of each component; a second extraction unit, configured to extract the spatial features based on the position information of each component and the local spatial relationship between adjacent components; and a third extraction unit, configured to extract the topological features based on the nesting relationship and connection relationship between different components, as well as the vertical relationship between the different components and the ground.

[0014] Optionally, in one embodiment of this application, the first construction module includes: a first generation unit, configured to convert the different types of components into graph nodes in the graph structure representation, and obtain the node attributes of the graph nodes based on the semantic features; a second generation unit, configured to obtain graph edges in the graph structure representation based on the spatial features or the topological features, and obtain the spatial edge attributes of the graph edges corresponding to the spatial features based on the local spatial relationships between adjacent components; an acquisition unit, configured to obtain the topological edge attributes of the graph edges corresponding to the topological features based on the nesting relationships, connection relationships, and vertical relationships in the topological features; and a third generation unit, configured to obtain the graph structure representation based on at least one of the graph nodes, the node attributes, the graph edges, the spatial edge attributes, and the topological edge attributes.

[0015] Optionally, in one embodiment of this application, the first training module includes: a fourth generation unit, configured to generate a corresponding occlusion dataset using the training dataset; a fifth generation unit, configured to input the occlusion dataset into the encoder of the pre-trained graph neural network model, so as to generate a first high-dimensional representation of the graph nodes and / or a second high-dimensional representation of the edge attributes in the occlusion dataset using the encoder; a sixth generation unit, configured to input the first high-dimensional representation and / or the second high-dimensional representation into the decoder of the pre-trained graph neural network model, so as to reconstruct the occluded graph nodes and occluded graph edges in the occlusion dataset using the decoder, so as to obtain reconstructed graph nodes and reconstructed graph edges; and a seventh generation unit, configured to obtain the trained training model based on the training dataset, the occlusion dataset, the reconstructed graph nodes, and the reconstructed graph edges.

[0016] A fourth aspect of this application provides a graph structure representation and error identification device for a BIM model, applied in the model transfer learning stage. The device includes: a second extraction module for extracting semantic features, spatial features, and topological features between different types of components based on component information of different types of components in the target BIM model; a generation module for generating a target graph structure representation of the target BIM model based on the semantic features, spatial features, and topological features; and an identification module for inputting the target graph structure representation into a pre-trained transfer learning model to identify erroneous target identification results of the target BIM model, wherein the pre-trained transfer learning model is obtained through transfer learning from a trained training model.

[0017] Optionally, in one embodiment of this application, it further includes: a third extraction module, configured to extract semantic features, spatial features, and topological features between different types of components in the BIM model containing the misidentified target based on component information of different types of components before inputting the target graph structure representation into the pre-trained transfer learning model; a second construction module, configured to generate an error graph structure representation of the BIM model containing the misidentified target based on the semantic features, spatial features, and topological features, and to construct a transfer dataset suitable for the pre-trained transfer learning model using the error graph structure representation; and a second training module, configured to train the pre-trained transfer learning model using the transfer dataset until a preset transfer condition is met, so as to obtain the trained transfer learning model.

[0018] A fifth aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. The processor executes the program to implement the graphical representation of a BIM model and the model error identification method as described in the above embodiments.

[0019] A sixth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described graphical representation of a BIM model and the method for identifying model errors.

[0020] A seventh aspect of this application provides a computer program product, including a computer program that, when executed, implements the graphical representation of the BIM model and the model error identification method described above.

[0021] This application's embodiments can extract semantic, spatial, and topological features between various types of components based on component information in a BIM model, thereby generating a graph structure representation of the BIM model. The graph structure representation is then used to construct a training dataset suitable for a pre-trained graph neural network model, which is then used to train the pre-trained graph neural network model. The trained model is then used for transfer learning to obtain a transfer learning model, which in turn identifies incorrectly identified targets in the BIM model. Through BIM component feature extraction, graph structure generation, pre-training, and transfer learning, the accuracy, generalization ability, and training efficiency of error identification can be significantly improved. This provides an intelligent and automated solution for BIM quality control, promoting the digital transformation of the construction industry. Therefore, it solves the problem in related technologies where the complex semantic, spatial, and topological features unique to BIM models pose significant challenges, making it difficult to perfectly apply pre-training and transfer learning models to BIM model features.

[0022] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0023] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0024] Figure 1 This is a flowchart illustrating a graphical representation of a BIM model and a method for identifying model errors, according to an embodiment of this application.

[0025] Figure 2 This is a schematic diagram illustrating the local spatial relationship between components according to an embodiment of this application;

[0026] Figure 3 This is a schematic diagram illustrating the topological relationships between components according to an embodiment of this application;

[0027] Figure 4 This is a partial flowchart of a graph structure representation provided according to an embodiment of this application;

[0028] Figure 5This is a schematic diagram of a GraphMAE2 graph neural network architecture provided according to an embodiment of this application;

[0029] Figure 6 This is a schematic diagram of an improved information transfer mechanism provided according to an embodiment of this application;

[0030] Figure 7 This is a block diagram illustrating the graphical representation of a BIM model and the model error identification device provided in the embodiments of this application.

[0031] Figure 8 This is a flowchart of a graphical representation of a BIM model and a method for identifying model errors according to yet another embodiment of this application;

[0032] Figure 9 This is a flowchart illustrating the working principle of a graphical representation of a BIM model and a model error identification method according to an embodiment of this application.

[0033] Figure 10 This is a schematic diagram illustrating three common types of misidentified targets in practical engineering, according to an embodiment of this application.

[0034] Figure 11 This is a block diagram of a graphical representation of a BIM model and a model error identification device according to yet another embodiment of this application;

[0035] Figure 12 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. Detailed Implementation

[0036] The embodiments of this application are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0037] The following description, with reference to the accompanying drawings, illustrates a graph structure representation of a BIM model and a method and apparatus for model error identification according to embodiments of this application. Addressing the challenges posed by the complex semantic, spatial, and topological features unique to BIM models, as mentioned in the background, and the inability to perfectly apply pre-training and transfer learning models to BIM model features, this application provides a graph structure representation of a BIM model and a method for model error identification. In this method, semantic, spatial, and topological features between various types of components can be extracted based on component information of different types of components in the BIM model, thereby generating a graph structure representation of the BIM model. This graph structure representation is then used to construct a training dataset suitable for a pre-trained graph neural network model, thereby training the pre-trained graph neural network model. The trained model is then used for transfer learning to obtain a transfer learning model, which in turn identifies error targets in the BIM model. Through BIM component feature extraction, graph structure generation, pre-training, and transfer learning, the accuracy, generalization ability, and training efficiency of error identification can be significantly improved, providing an intelligent and automated solution for BIM quality control and promoting the digital transformation of the construction industry. This solves the problem that the complex semantic, spatial, and topological features unique to BIM models pose significant challenges in related technologies, making them unsuitable for pre-training and transfer learning models of BIM model features.

[0038] Specifically, Figure 1 This is a flowchart of a graphical representation of a BIM model and a method for identifying model errors, provided according to an embodiment of this application.

[0039] like Figure 1 As shown, the graph structure representation of the BIM model and the method for model error identification are applied in the model training phase. The method includes the following steps:

[0040] In step S101, semantic features, spatial features, and topological features among different types of components are extracted based on the component information of different types of components in the BIM model.

[0041] It is understood that the embodiments of this application can realize a unified vectorized representation of BIM models: extract and calculate the "semantic-spatial-topological" multidimensional features between various types of components in the BIM model based on the component information of different types of components in the BIM model.

[0042] Furthermore, in the embodiments of this application, all components can be represented as nodes of different types according to their corresponding types. The extracted semantic features (such as geometric properties, material information, etc., which are not specifically limited in this application) are represented as node features through numerical values, One-Hot encoding, or multilingual large-scale embedding model encoding. The spatial and topological features between components can be represented as edges connecting the corresponding nodes. The corresponding spatial relationships (such as the shortest distance between two components, the included angle, etc., which are not specifically limited in this application) and topological relationships (such as connection relationships, embedding relationships, etc., which are not specifically limited in this application) are represented as corresponding edge attributes through numerical values ​​or One-Hot encoding. The specific representation method can be set by those skilled in the art according to the actual situation, and this application does not impose any specific restrictions.

[0043] As one possible approach, embodiments of this application can extract semantic features, spatial features, and topological features between various types of components based on component information of different types of components in the BIM model.

[0044] Optionally, in one embodiment of this application, semantic features, spatial features, and topological features between different types of components are extracted based on the component information of different types of components in the BIM model, including: extracting semantic features based on the component feature information of each component; extracting spatial features based on the location information of each component and the local spatial relationship between adjacent components; and extracting topological features based on the nesting relationship and connection relationship between different components, as well as the vertical relationship between different components and the ground.

[0045] In some embodiments, the present application can extract component feature information of different components based on the IFC (Industry Foundation Classes) file corresponding to the BIM model, and then obtain the semantic features of various components in the BIM model.

[0046] In this application embodiment, semantic features may include the shape, size features, structural purpose, family and family type of the component extracted from the IFC file, etc. This application does not impose specific limitations. It may also include the features of each component itself, such as basic information, geometric shape, material properties, other related attributes, etc. This application does not impose specific limitations.

[0047] Furthermore, in the embodiments of this application, the shape of the component may include, but is not limited to, cubes, cylinders, and other irregular shapes, etc., and this application does not impose specific limitations; the dimensional features may include, but are not limited to, the three dimensions of length, width, and height, etc., and this application does not impose specific limitations. It should be noted that, in the embodiments of this application, for cylindrical components (e.g., pipes, this application does not impose specific limitations), the width and height can be represented by the cross-sectional radius; for irregularly shaped components (e.g., certain connectors, this application does not impose specific limitations), three main dimensional parameters that can represent their geometric features can be selected, and this application does not impose specific limitations; the structural purpose can be... The functional positioning of components can be characterized, but is not limited to, structural and non-structural components; this application does not impose specific limitations. Families and family types can be used to characterize the functional purpose of an object and its detailed attributes such as geometric dimensions and materials; this application does not impose specific limitations. Basic information can include, but is not limited to, attributes such as component ID (Identifier), name, and family name; this application does not impose specific limitations and is used to uniquely identify components. Geometric shapes can include, but are not limited to, bounding boxes and cross-sectional shapes; this application does not impose specific limitations. Material properties can describe the composition and characteristics of components; this application does not impose specific limitations. Other related attributes can include, but are not limited to, durability, thermal conductivity, and compressive strength; this application does not impose specific limitations.

[0048] In some embodiments, the present application can extract spatial features based on the location information of each component and the local spatial relationship between adjacent components.

[0049] In this embodiment, the location information representing the component's position, namely the local coordinate system and the reference coordinate system, can be extracted from the IFC file corresponding to the BIM model. Then, the local spatial relationship between a component and its other neighboring components within a 0.5m range can be calculated to provide additional geometric and contextual information.

[0050] For example, the embodiments of this application are combined with Figure 2 As shown, the local spatial relationships between different components are given. These local spatial relationships may include, but are not limited to, spatial relationships, included angles, shortest distance vector coordinates, shortest distance, and the angle between the plane formed by the centerline and the horizontal plane. This application does not impose specific limitations.

[0051] Furthermore, in the embodiments of this application, the spatial relationship can be understood as the positional relationship between two components in space, determined by their center point or center line. It may include, but is not limited to: (1) skew planes; (2) coplanar but not parallel; (3) coplanar and parallel; (4) point-to-line; (5) point-to-point, etc., and this application does not impose specific limitations. Among them, (1), (2), and (3) are applicable to the case where both components are positioned using center lines; (4) is applicable to the case where one component is positioned using a center point and the other component is positioned using a center line; (5) is applicable to the case where both components are positioned using center points.

[0052] An included angle can be understood as the angle between the center lines or points of two components. If a component is located using a center point, then the angle can be considered as 0°.

[0053] The coordinates of the shortest distance vector can correspond to the coordinates of the shortest distance vector between the center points or center lines of two components.

[0054] The shortest distance can represent the shortest distance between two components. A negative distance indicates overlap or collision, while a positive distance indicates the shortest distance between components.

[0055] The angle between the plane formed by the centerline and the horizontal plane can be understood as the angle between the plane or line formed by the two components and the horizontal plane. For the spatial relationships in (2)-(5), the angle between the plane formed by the centerline and the horizontal plane is the angle between the plane or line formed by the center point or line and the horizontal plane. For the spatial relationship in (1), this angle is 0°. According to geometric principles, it is clear that these five calculated spatial relationships can comprehensively describe the relative positional relationship between the two components.

[0056] In some embodiments, the embodiments of this application can extract topological features based on the nesting relationship and connection relationship between different components, as well as the vertical relationship between different components and the ground.

[0057] In this embodiment, the nesting and connection relationships contained in the IFC file corresponding to the BIM model can be extracted, and the vertical relationship between the component and the ground can be calculated, thereby extracting the corresponding topological features.

[0058] For example, the embodiments of this application are combined with Figure 3 As shown, the topological relationships between different components are given. These topological relationships may include, but are not limited to, nesting relationships, connection relationships, and vertical relationships between components and the ground (contact floor slab).

[0059] Nesting can be understood as a construction situation where one component is completely nested inside another component.

[0060] Connection relationship can be understood as a direct physical connection between two components or the requirement for components to maintain continuity in design semantics.

[0061] The contact slab can represent the contact relationship between the component and the ground. This topological relationship can be extracted by calculating the positional relationship between the component and the floor boundary frame, so as to more intuitively represent the functional layout and component distribution of different building floors.

[0062] In step S102, a graph structure representation of the BIM model is generated based on semantic features, spatial features, and topological features, and a training dataset suitable for pre-trained graph neural network models is constructed using the graph structure representation.

[0063] In actual implementation, the embodiments of this application can obtain the graph structure representation of the BIM model based on semantic features, spatial features and topological features, and then use the graph structure representation to construct a training dataset suitable for pre-trained graph neural network models.

[0064] Optionally, in one embodiment of this application, generating a graph structure representation of the BIM model based on semantic features, spatial features, and topological features includes: converting different types of components into graph nodes in the graph structure representation, and obtaining node attributes of the graph nodes based on semantic features; obtaining graph edges in the graph structure representation based on spatial features or topological features, and obtaining spatial edge attributes of graph edges corresponding to spatial features based on local spatial relationships between adjacent components; obtaining topological edge attributes of graph edges corresponding to topological features based on nesting relationships, connection relationships, and vertical relationships in topological features; and obtaining a graph structure representation based on at least one of graph nodes, node attributes, graph edges, spatial edge attributes, and topological edge attributes.

[0065] In some embodiments, all different types of components can be represented as different types of graph nodes, and the extracted semantic features can be represented as the node attributes of the corresponding graph nodes.

[0066] For example, in the embodiments of this application, all components in all BIM models can be regarded as graph nodes in the graph structure representation; and based on the component type classification in IFC, these graph nodes are divided into different types, such as walls, beams, pipes, etc. This application does not make specific limitations, and then the extracted semantic features are converted into the node attributes of the corresponding graph nodes.

[0067] In some embodiments, the spatial and topological relationships between components extracted and calculated can be represented as graph edges connecting corresponding nodes, thereby obtaining the corresponding spatial edge attributes and topological edge attributes.

[0068] For example, embodiments of this application can further classify graph edges representing topological features based on nesting, connection, and vertical relationships in the topological features, thereby obtaining the topological edge attributes of the graph edges corresponding to the topological features; spatial features between components are regarded as the fourth type of graph edges in the graph structure representation, and local spatial relationships between adjacent components are used as spatial edge attributes of the graph edges corresponding to the spatial features; when there are topological features between two components, their spatial features are ignored to ensure that the two component nodes are connected by only one edge.

[0069] In addition, it should be noted that the embodiments of this application can use various embedding techniques to vectorize node features and align and fill missing edge attributes, thereby realizing graph-based representation of BIM models and forming a training dataset for pre-trained graph neural network models.

[0070] For example, embodiments of this application can encode the extracted non-numerical node attributes to convert them into a computer-readable form; for certain components that lack specific node and edge attributes or whose values ​​cannot be extracted, 0 placeholders are used to maintain consistency; all node and edge attributes are normalized, as shown in the flowchart below. Figure 4 As shown, the content can be:

[0071] Step S401: Encode the extracted non-numeric node attributes to convert them into a computer-readable form.

[0072] The encoding methods used in the embodiments of this application may include, but are not limited to, One-hot encoding, Numeric encoding, and M3E (Moka Massive Mixed Embedding) encoding, etc., and this application does not impose specific limitations. One-hot encoding can be understood as representing categorical variables as sparse vectors, where only the positions corresponding to the category are 1, and the remaining positions are 0; M3E encoding can be understood as a method from a library that embeds text into a fixed-dimensional (e.g., 64-dimensional, this application does not impose specific limitations) vector representation, facilitating model training and inference.

[0073] Furthermore, the encoding methods for each attribute provided in the embodiments of this application are shown in Table 1. Table 1 is a schematic table illustrating the encoding methods for each attribute according to one embodiment of this application.

[0074] Table 1

[0075]

[0076] Step S402: For certain components that lack specific node and edge attributes or whose values ​​cannot be extracted, use 0 placeholders to maintain consistency.

[0077] Step S403: Normalize the attributes of all nodes and edges.

[0078] In this embodiment, the normalization process is performed on a per-floor drawing basis, scaling the relevant attribute values ​​to a range of -1 to 1 while keeping the 0 value unchanged. The calculation formula can be, but is not limited to, expressed as:

[0079]

[0080] Among them, X i X′ represents the i-th attribute value of all components in the same floor plan. i This represents the normalized data.

[0081] In step S103, a pre-trained graph neural network model is trained using the training dataset to obtain a trained model, and the trained model is used for transfer learning to obtain a transfer learning model, which is then used to identify incorrectly identified targets in the BIM model.

[0082] It is understood that the embodiments of this application can utilize a pre-trained graph neural network model to learn the rich design experience and rules implicit in the BIM model, and use the training dataset to train the pre-trained graph neural network model, thereby obtaining the trained model.

[0083] Pre-training is a strategy for training deep learning models. Its core lies in using large-scale datasets to initially train the model, enabling it to learn general feature representations. This process is similar to the basic learning stage humans undergo before learning new knowledge, accumulating experience through extensive reading and observation.

[0084] In some embodiments, the present application can utilize the trained training model to perform transfer learning, thereby obtaining a corresponding transfer learning model, and use the transfer learning model to identify incorrectly identified targets in the BIM model.

[0085] Optionally, in one embodiment of this application, training a pre-trained graph neural network model using a training dataset to obtain a trained model includes: generating a corresponding occlusion dataset using the training dataset; inputting the occlusion dataset into the encoder of the pre-trained graph neural network model to generate a first high-dimensional representation of the graph nodes and / or a second high-dimensional representation of the edge attributes in the occlusion dataset; inputting the first high-dimensional representation and / or the second high-dimensional representation into the decoder of the pre-trained graph neural network model to reconstruct the occluded graph nodes and occluded graph edges in the occlusion dataset using the decoder to obtain reconstructed graph nodes and reconstructed graph edges; and obtaining a trained model based on the training dataset, the occlusion dataset, the reconstructed graph nodes, and the reconstructed graph edges.

[0086] It is understood that, in the embodiments of this application, the pre-trained graph neural network model may include two parts: an encoder and a decoder. The encoder's task is to take the input graph in which some nodes and edges are occluded, and construct a first high-dimensional representation of the graph nodes and a second high-dimensional representation of the edge attributes, which contain global features, by capturing the features of the local structure of the graph, i.e., the unoccluded nodes and edges and their neighborhoods. The decoder then reconstructs and recovers the occluded nodes and edge attributes in the input graph based on the first and second high-dimensional representations, thereby obtaining the reconstructed graph nodes and reconstructed graph edges.

[0087] In some embodiments, the present application can use a training dataset to train a pre-trained graph neural network model, thereby obtaining a trained model.

[0088] For example, in this embodiment of the application, the training dataset can be input into a pre-trained graph neural network model. By masking some node and edge features in the input graph, the pre-training problem is transformed into a problem of decoding and reconstructing node and edge features. This prompts the model to self-supervised learn and recover the entire graph structure from local information, thereby obtaining the trained graph neural network model. The architecture selected for the pre-trained graph neural network model is the GraphMAE2 graph neural network architecture, as illustrated in the diagram below. Figure 5 As shown, the architecture consists of two parts: an encoder and a decoder. The purpose of pre-training is to obtain a trained encoder by inputting the reconstructed graph nodes and edges. The main task of the encoder is to construct a first high-dimensional representation of the graph nodes containing global features and a second high-dimensional representation of the edge attributes by capturing the local structure of the graph, i.e., the features of the unmasked nodes and their neighborhoods, from the partially occluded graph. The decoder then reconstructs and recovers the occluded nodes and edge attributes in the input graph based on the first and second high-dimensional representations. It includes two decoding strategies: (1) a multi-view random remasking strategy to reduce overfitting of the input features; and (2) latent representation prediction to obtain more information targets. The specific settings can be made by those skilled in the art according to the actual situation. This application does not impose any specific restrictions.

[0089] Furthermore, in this embodiment, the multi-view random remasking strategy can be understood as randomly remasking the embedded high-dimensional features during decoding, i.e., randomly masking different nodes multiple times, and reconstructing the input features from the hidden codes of unmasked neighboring nodes using a shared decoder. The introduction of randomness serves as regularization, thereby reducing the sensitivity of training to perturbations in the input features and reducing overfitting. After hyperparameter tuning, this embodiment uses a large masking rate (50%) during both encoding and decoding. The attributes of the masked nodes are randomly filled. During decoding, three remasked views are randomly generated for decoding, and the scaling cosine error is used to measure the reconstruction error. The sum of the errors from the three views is used for training. The calculation process can be, but is not limited to, expressed as:

[0090]

[0091] Where V represents the input mask node, K represents the number of random remasking operations during decoding (which can be 3, and this application does not impose a specific restriction), and x i This represents the i-th feature of the input. Z represents the predicted feature matrix under the j-th masked view. (j) The i-th row, cos(θ) ij ) represents two vectors x i and The angle between them, T represents transpose, v i This represents one of the input nodes.

[0092] The core of latent representation prediction lies in designing an additional information prediction task while minimizing the direct impact of input features on this task. Therefore, embodiments of this application can choose to perform prediction in the representation space rather than the input feature space. This part involves three networks: a target generator, an encoder, and an MLP (Multilayer Perceptron) projector. The target generator's role is to generate latent predicted targets from the unmasked image. It shares the same network architecture as the encoder and MLP projector, but uses different weights. Furthermore, embodiments of this application can project the encoder's encoding result onto the representation space using the MLP projector to obtain latent predictions. By minimizing and The encoder is trained using the distance between them, and the calculation process can be, but is not limited to, represented as...

[0093]

[0094] ζ←τζ+(1-τ)ξ,

[0095] Where N represents the number of nodes. It is a matrix The i-th row, It is a matrix The i-th row, The angle between the two vectors is represented by the learnable weight ζ in the target generator, which is updated by an exponential moving average of the learnable weights ξ of the encoder and MLP projector and using the weight decay parameter τ.

[0096] Furthermore, the embodiments of this application can yield the final loss function, the expression of which may be, but is not limited to, the following:

[0097]

[0098] It should be noted that, in the embodiments of this application, both the encoder and the decoder are two-layer graph attention networks.

[0099] Figure 5 The GraphMAE2 graph neural network architecture shown is only applicable to homogeneous graphs and not to heterogeneous networks. Therefore, the embodiments of this application can improve the network information transfer mechanism of this architecture to make it applicable to heterogeneous networks. Figure 6 This is a schematic diagram of the improved information transfer mechanism, such as... Figure 6 As shown, in this embodiment of the application, neighbor information can be processed differently according to node type, and semantic, spatial and topological node features can be aggregated by introducing a type-aware attention mechanism.

[0100] Furthermore, embodiments of this application can input the training dataset to... Figure 6 In the improved GraphMAE2 graph neural network shown, the training problem is transformed into a problem of node feature decoding and reconstruction by masking some node and edge features in the input graph. This prompts the model to learn in a self-supervised manner to recover the entire graph structure from local information, thereby obtaining the trained model.

[0101] The graph representation and error identification method for BIM models proposed in this application can extract semantic, spatial, and topological features between different types of components based on component information in the BIM model, thereby generating a graph representation of the BIM model. This graph representation is then used to construct a training dataset suitable for a pre-trained graph neural network model, which is then used to train the pre-trained model. The trained model is then used for transfer learning to obtain a transfer learning model, which is used to identify error targets in the BIM model. Through BIM component feature extraction, graph generation, pre-training, and transfer learning, the accuracy, generalization ability, and training efficiency of error identification can be significantly improved, providing an intelligent and automated solution for BIM quality control and promoting the digital transformation of the construction industry. This solves the problem in related technologies where the complex semantic, spatial, and topological features unique to BIM models pose significant challenges, making it difficult to perfectly apply pre-training and transfer learning models to BIM model features.

[0102] Next, referring to the accompanying drawings, we describe the graphical representation of the BIM model and the model error identification device proposed according to the embodiments of this application.

[0103] Figure 7 This is a block diagram of the graphical representation of a BIM model and the model error identification device provided in the embodiments of this application.

[0104] like Figure 7 As shown, the BIM model graph structure representation and model error identification device 10 is applied in the model training stage. The BIM model graph structure representation and model error identification device 10 includes: a first extraction module 100, a first construction module 200 and a first training module 300.

[0105] The first extraction module 100 is used to extract semantic features, spatial features and topological features between different types of components based on the component information of different types of components in the BIM model.

[0106] The first building module 200 is used to generate a graph structure representation of the BIM model based on semantic features, spatial features and topological features, and to use the graph structure representation to build a training dataset suitable for pre-trained graph neural network models.

[0107] The first training module 300 is used to train a pre-trained graph neural network model using the training dataset to obtain a trained model, and to perform transfer learning using the trained model to obtain a transfer learning model, which is then used to identify incorrectly identified targets in the BIM model.

[0108] Optionally, in one embodiment of this application, the first extraction module 100 includes: a first extraction unit, a second extraction unit, and a third extraction unit.

[0109] The first extraction unit is used to extract semantic features based on the component feature information of each component.

[0110] The second extraction unit is used to extract spatial features based on the location information of each component and the local spatial relationship between neighboring components.

[0111] The third extraction unit is used to extract topological features based on the nesting and connection relationships between different components, as well as the vertical relationship between different components and the ground.

[0112] Optionally, in one embodiment of this application, the first construction module 200 includes: a first generation unit, a second generation unit, an acquisition unit, and a third generation unit.

[0113] The first generation unit is used to convert different types of components into graph nodes in the graph structure representation, and obtain the node attributes of the graph nodes based on semantic features.

[0114] The second generation unit is used to obtain graph edges in the graph structure representation based on spatial features or topological features, and to obtain the spatial edge attributes of the graph edges corresponding to the spatial features based on the local spatial relationships between neighboring components.

[0115] The unit is obtained by using nested, connected, and vertical relationships in the topological features to obtain the topological edge attributes of the graph edges corresponding to the topological features.

[0116] The third generation unit is used to obtain a graph structure representation based on at least one of graph nodes, node attributes, graph edges, spatial edge attributes, and topological edge attributes.

[0117] Optionally, in one embodiment of this application, the first training module 300 includes: a fourth generation unit, a fifth generation unit, a sixth generation unit, and a seventh generation unit.

[0118] The fourth generation unit is used to generate the corresponding occlusion dataset using the training dataset.

[0119] The fifth generation unit is used to input the occluded dataset into the encoder of the pre-trained graph neural network model to generate a first high-dimensional representation of the graph nodes and / or a second high-dimensional representation of the edge attributes in the occluded dataset using the encoder.

[0120] The sixth generation unit is used to input the first high-dimensional representation and / or the second high-dimensional representation into the decoder of the pre-trained graph neural network model, so as to use the decoder to reconstruct the occluded graph nodes and occluded graph edges in the occluded dataset, so as to obtain the reconstructed graph nodes and reconstructed graph edges.

[0121] The seventh generation unit is used to obtain the trained model based on the training dataset, the occlusion dataset, the reconstructed graph nodes, and the reconstructed graph edges.

[0122] It should be noted that the foregoing explanation of the BIM model graphical representation and model error identification method embodiment also applies to the BIM model graphical representation and model error identification device 10 of this embodiment, and will not be repeated here.

[0123] The BIM model graph structure representation and model error identification device 10 proposed in this application can extract semantic features, spatial features, and topological features between various types of components based on the component information of different types of components in the BIM model, thereby generating a graph structure representation of the BIM model. The graph structure representation is then used to construct a training dataset suitable for a pre-trained graph neural network model, thereby training the pre-trained graph neural network model. The trained model is then used for transfer learning to obtain a transfer learning model, which in turn identifies error identification targets in the BIM model. Through BIM component feature extraction, graph structure generation, pre-training, and transfer learning, the accuracy, generalization ability, and training efficiency of error identification can be significantly improved, providing an intelligent and automated solution for BIM quality control and promoting the digital transformation of the construction industry. This solves the problem in related technologies where the complex semantic, spatial, and topological features unique to BIM models pose significant challenges, making it difficult to perfectly apply pre-training and transfer learning models for BIM model features.

[0124] The above embodiments describe the model training phase. The following describes embodiments for the model transfer learning phase.

[0125] Figure 8 This is a flowchart of a graphical representation of a BIM model and a method for identifying model errors, according to yet another embodiment of this application.

[0126] like Figure 8 As shown, the graph structure representation and error identification method of the BIM model are applied to the model transfer learning stage. The method includes the following steps:

[0127] In step S801, semantic features, spatial features, and topological features among different types of components are extracted based on the component information of different types of components in the target BIM model.

[0128] In step S802, a target graph structure representation of the target BIM model is generated based on semantic features, spatial features, and topological features.

[0129] In step S803, the target graph structure representation is input into a pre-trained transfer learning model to identify erroneous target identification results of the target BIM model, wherein the pre-trained transfer learning model is obtained by transfer learning from the trained model.

[0130] Optionally, in one embodiment of this application, before inputting the target graph structure representation into the pre-trained transfer learning model, the method further includes: extracting semantic features, spatial features, and topological features between different types of components based on component information of different types of components in the BIM model containing the misidentified target; generating an error graph structure representation of the BIM model containing the misidentified target based on the semantic features, spatial features, and topological features, and constructing a transfer dataset suitable for the pre-trained transfer learning model using the error graph structure representation; and training the pre-trained transfer learning model using the transfer dataset until a preset transfer condition is met to obtain the trained transfer learning model.

[0131] In some embodiments, the present application can represent the target BIM model as a corresponding target graph structure representation. A small number of BIM model data represented as incorrect graph structures are labeled based on the incorrectly identified targets, thereby forming a transfer dataset suitable for a pre-trained transfer learning model. The pre-trained transfer learning model is then fine-tuned using the transfer learning dataset until certain transfer conditions are met. For example, by inputting the target BIM model with the incorrect graph structure representation removed into the trained transfer learning model, automatic error identification of BIM models under few-sample transfer can be achieved. These certain transfer conditions can be set by those skilled in the art according to actual circumstances, and the present application does not impose specific limitations.

[0132] Furthermore, the embodiment of this application generates an error graph structure representation of the BIM model containing the error identification target, and uses the error graph structure representation to construct a transfer dataset suitable for the pre-trained transfer learning model. The content can be as follows: the error identification task in the BIM model containing the error identification target is converted into a node and edge classification problem; a small amount of data to be identified is labeled with correct nodes or edges as 1 and incorrect nodes or edges as 0 according to the error identification target, thereby forming a transfer dataset for transfer learning.

[0133] In addition, the content of training the pre-trained transfer learning model using the transfer dataset in this application embodiment can be as follows: only the encoder in the pre-trained transfer learning model is retained for feature extraction, and a new graph attention network is connected after the encoder as a classifier; the encoder is kept unchanged during training, and only the classifier is trained and optimized, thereby greatly improving the computational accuracy and supporting the unified identification of errors in various BIM models, which is something that existing methods cannot do.

[0134] The working principle of the graphical representation of the BIM model and the model error identification method proposed in this application will be explained below with reference to a specific embodiment.

[0135] in, Figure 9 This is a flowchart illustrating the working principle of a graphical representation of a BIM model and a method for identifying model errors according to an embodiment of this application.

[0136] Step S901: Represent the target BIM model as a corresponding target graph structure representation.

[0137] In this embodiment, semantic features, spatial features, and topological features between different types of components can be extracted based on the component information of different types of components in the target BIM model, and then the corresponding target graph structure representation can be obtained by using the semantic features, spatial features, and topological features.

[0138] Step S902: Generate an error graph structure representation of the BIM model containing the incorrectly identified targets, thereby obtaining the corresponding migration dataset.

[0139] In this application, the embodiments are combined with Figure 10 Three common error identification targets in actual engineering were selected: (1) Semantic errors: For example, some walls were incorrectly created using beam families, which led to a chain reaction in subsequent cost estimation, operation and maintenance, etc., resulting in deviations in cost estimation and increased maintenance difficulty; (2) Attribute value errors: For example, the height of some doors did not meet the relevant requirements. According to industry standards, the height of a door should generally not be less than 200 cm and should not exceed 240 cm; (3) Topological relationship errors: For example, the connection relationship between some electromechanical pipeline components and their connectors in terms of distance or attributes was missing. This usually occurs when modifying drawings and moving components. The position of the electromechanical pipeline components or their corresponding connectors changes, resulting in gaps between them. Even if they are geometrically connected, the connection in terms of relative movement attributes may disappear. The specific selection method can be set by those skilled in the art according to the actual situation. This application does not impose specific restrictions.

[0140] Furthermore, in this embodiment of the application, a small amount of data to be identified can be labeled with 1 for correct nodes or edges and 0 for incorrect nodes or edges according to three types of error identification targets, thereby forming a transfer learning dataset.

[0141] Step S903: Fine-tune the pre-trained transfer learning model using the transfer learning dataset.

[0142] In this embodiment, only the encoder in the pre-trained transfer learning model is retained for feature extraction, and a new graph attention network with the same architecture is connected after the encoder as a classifier; during training, the encoder is kept unchanged, and only the classifier is trained and optimized. Figure 10 The model, after transfer learning, can classify nodes and edges into four categories based on the three error types: correct, error 1, error 2, and error 3.

[0143] Step S904: Input the target graph structure representation into the pre-trained transfer learning model to identify erroneous target identification results of the target BIM model.

[0144] The graph structure representation and error identification method for BIM models proposed in this application can extract semantic, spatial, and topological features between various types of components based on component information in the target BIM model. This generates a target graph structure representation of the target BIM model, which is then input into a pre-trained transfer learning model to identify erroneous target identification results. Through BIM component feature extraction, graph structure generation, pre-training, and transfer learning, the accuracy, generalization ability, and training efficiency of error identification can be significantly improved. This provides an intelligent and automated solution for BIM quality control, promoting the digital transformation of the construction industry. Therefore, it solves the problem in related technologies where the complex semantic, spatial, and topological features unique to BIM models pose significant challenges, making it difficult to perfectly apply to pre-training and transfer learning models for BIM model features.

[0145] Next, referring to the accompanying drawings, we describe the graphical representation of the BIM model and the model error identification device proposed according to the embodiments of this application.

[0146] Figure 11 A block diagram illustrating the graphical representation of a BIM model and a model error identification device, as provided in another embodiment of this application.

[0147] like Figure 11 As shown, the BIM model graph structure representation and model error identification device 20 is applied in the model transfer learning stage. The BIM model graph structure representation and model error identification device 20 includes: a second extraction module 400, a generation module 500 and an identification module 600.

[0148] The second extraction module 400 is used to extract semantic features, spatial features and topological features between different types of components based on the component information of different types of components in the target BIM model.

[0149] The generation module 500 is used to generate a target graph structure representation of the target BIM model based on semantic features, spatial features, and topological features.

[0150] The recognition module 600 is used to input the target graph structure representation into a pre-trained transfer learning model to identify erroneous target recognition results of the target BIM model, wherein the pre-trained transfer learning model is obtained by transfer learning from the trained model.

[0151] Optionally, in one embodiment of this application, it further includes: a third extraction module, a second construction module, and a second training module.

[0152] The third extraction module is used to extract semantic features, spatial features, and topological features among different types of components based on component information of different types of components in the BIM model containing misidentified targets before inputting the target graph structure representation into the pre-trained transfer learning model.

[0153] The second building module is used to generate an error graph structure representation of the BIM model containing the incorrectly identified target based on semantic features, spatial features, and topological features, and to use the error graph structure representation to build a transfer dataset suitable for pre-trained transfer learning models.

[0154] The second training module is used to train a pre-trained transfer learning model using the transfer dataset until the preset transfer conditions are met, so as to obtain the trained transfer learning model.

[0155] It should be noted that the foregoing explanation of the BIM model graphical representation and model error identification method embodiment also applies to the BIM model graphical representation and model error identification device 20 of this embodiment, and will not be repeated here.

[0156] The BIM model graph structure representation and model error identification device 20 proposed in this application can extract semantic features, spatial features, and topological features between various types of components based on the component information of different types of components in the target BIM model, thereby generating a target graph structure representation of the target BIM model. This target graph structure representation is then input into a pre-trained transfer learning model to identify erroneous target identification results in the target BIM model. Through BIM component feature extraction, graph structure generation, pre-training, and transfer learning, the accuracy, generalization ability, and training efficiency of error identification can be significantly improved, providing an intelligent and automated solution for BIM quality control and promoting the digital transformation of the construction industry. This solves the problem in related technologies where the complex semantic, spatial, and topological features unique to BIM models pose significant challenges, making it difficult to perfectly apply to pre-training and transfer learning models of BIM model features.

[0157] Figure 12 This is a schematic diagram of the structure of an electronic device provided according to an embodiment of this application. The electronic device may include:

[0158] The memory 1201, the processor 1202, and the computer program stored on the memory 1201 and executable on the processor 1202.

[0159] When the processor 1202 executes the program, it implements the graphical representation of the BIM model and the model error identification method provided in the above embodiments.

[0160] Furthermore, electronic devices also include:

[0161] Communication interface 1203 is used for communication between memory 1201 and processor 1202.

[0162] The memory 1201 is used to store computer programs that can run on the processor 1202.

[0163] The memory 1201 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0164] If the memory 1201, processor 1202, and communication interface 1203 are implemented independently, then the communication interface 1203, memory 1201, and processor 1202 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 12 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0165] Optionally, in a specific implementation, if the memory 1201, processor 1202, and communication interface 1203 are integrated on a single chip, then the memory 1201, processor 1202, and communication interface 1203 can communicate with each other through an internal interface.

[0166] The processor 1202 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0167] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for graphical representation of the BIM model and model error identification.

[0168] This application also provides a computer program product, including a computer program that, when executed, implements the above-described graphical representation of the BIM model and the model error identification method.

[0169] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0170] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0171] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0172] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). In addition, computer-readable media can even be paper or other suitable media on which programs can be printed, because programs can be obtained electronically by optically scanning paper or other media, followed by editing, interpreting or otherwise processing as necessary, and then stored in computer memory.

[0173] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or more of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0174] Those skilled in the art will understand that all or part of the steps of the methods described in the above embodiments can be implemented by a program instructing related hardware, and the program can be stored in a computer-readable storage medium. When executed, the program includes one or a combination of the steps of the method embodiments.

[0175] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0176] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for graph structure representation of BIM model and model error identification, characterized in that, The method is applied to a model training stage, and the method comprises the following steps: According to the component information of different types of components in the building information model (BIM) model, the semantic features, spatial features and topological features between various components are extracted; Based on the semantic features, spatial features and topological features, a graph structure representation of the BIM model is generated, and a training data set suitable for a pre-trained graph neural network model is constructed using the graph structure representation; The pre-trained graph neural network model is trained using the training data set to obtain a trained training model, and the trained training model is used for transfer learning to obtain a transfer learning model, so that the transfer learning model is used to identify error recognition targets of the BIM model; Wherein, the graph structure representation of the BIM model is generated based on the semantic features, spatial features and topological features, comprising: The different types of components are converted into graph nodes in the graph structure representation, and the node attributes of the graph nodes are obtained based on the semantic features; Based on the spatial features or the topological features, the graph edges in the graph structure representation are obtained, and the spatial edge attributes of the spatial features corresponding to the graph edges are obtained based on the local spatial relationship between adjacent components; Based on the nesting relationship, connection relationship and vertical relationship in the topological features, the topological edge attributes of the topological features corresponding to the graph edges are obtained; Based on at least one of the graph nodes, node attributes, graph edges, spatial edge attributes and topological edge attributes, the graph structure representation is obtained; The pre-trained graph neural network model is trained using the training data set to obtain a trained training model, comprising: A corresponding masking data set is generated using the training data set; The masking data set is input into an encoder of the pre-trained graph neural network model to generate first high-dimensional representations of graph nodes and / or second high-dimensional representations of edge attributes in the masking data set using the encoder; The first high-dimensional representations and / or the second high-dimensional representations are input into a decoder of the pre-trained graph neural network model to reconstruct the masked graph nodes and the masked graph edges in the masking data set using the decoder to obtain reconstructed graph nodes and reconstructed graph edges; Based on the training data set, the masking data set, the reconstructed graph nodes and the reconstructed graph edges, the trained training model is obtained.

2. The method of claim 1, wherein, The semantic features, spatial features and topological features between various components are extracted according to the component information of different types of components in the BIM model, comprising: Based on the component feature information of each component, the semantic features are extracted; Based on the position information of each component and the local spatial relationship between adjacent components, the spatial features are extracted; Based on the nesting relationship, connection relationship between different components, and the vertical relationship between the different components and the ground, the topological features are extracted.

3. A method for graph structure representation of BIM model and model error identification, characterized in that, The model training stage adopts the graph structure representation of the BIM model and the model error recognition method according to any one of claims 1-2, and is applied to a model transfer learning stage, wherein the method comprises the following steps: According to the component information of different types of components in the target BIM model, semantic features, spatial features and topological features between components of various types are extracted; Based on the semantic features, spatial features and topological features, a target graph structure representation of the target BIM model is generated; The target graph structure representation is input into a pre-trained transfer learning model to identify the error target recognition result of the target BIM model, wherein the pre-trained transfer learning model is obtained by transfer learning of the trained training model.

4. The method of claim 3, wherein, Before inputting the target graph structure representation into the pre-trained transfer learning model, it further includes: According to the component information of different types of components in the BIM model containing the error recognition target, semantic features, spatial features and topological features between components of various types are extracted; Based on the semantic features, spatial features and topological features, an error graph structure representation of the BIM model containing the error recognition target is generated, and a transfer data set suitable for the pre-training transfer learning model is constructed using the error graph structure representation; The pre-trained transfer learning model is trained using the transfer data set until the preset transfer condition is met to obtain the trained transfer learning model.

5. A device for graph structure representation of a BIM model and model error identification, characterized in that, Applied to the model training stage, wherein the device includes: A first extraction module is configured to extract semantic features, spatial features and topological features between components of various types according to component information of different types of components in a BIM model; A first construction module is configured to generate a graph structure representation of the BIM model based on the semantic features, spatial features and topological features, and to construct a training data set suitable for a pre-trained graph neural network model using the graph structure representation; A first training module is configured to train the pre-trained graph neural network model using the training data set to obtain a trained training model, and to perform transfer learning using the trained training model to obtain a transfer learning model, so as to identify an error recognition target of the BIM model using the transfer learning model; The first construction module includes: A first generation unit is configured to convert the different types of components into graph nodes in the graph structure representation, and to obtain node attributes of the graph nodes based on the semantic features; A second generation unit is configured to obtain graph edges in the graph structure representation based on the spatial features or the topological features, and to obtain spatial edge attributes of the spatial features corresponding graph edges based on local spatial relationships between adjacent components; An acquisition unit is configured to obtain topological edge attributes of the topological features corresponding graph edges based on nested relationships, connection relationships and vertical relationships in the topological features; A third generation unit is configured to obtain the graph structure representation based on at least one of the graph nodes, the node attributes, the graph edges, the spatial edge attributes and the topological edge attributes; The first training module includes: A fourth generation unit is configured to generate a corresponding masking data set using the training data set; The fifth generation unit is configured to input the masking data set into an encoder of the pre-trained graph neural network model to generate first high-dimensional representations of graph nodes and / or second high-dimensional representations of edge attributes in the masking data set by using the encoder. The sixth generation unit is configured to input the first high-dimensional representations and / or the second high-dimensional representations into a decoder of the pre-trained graph neural network model to reconstruct the masked graph nodes and the masked graph edges in the masking data set by using the decoder to obtain reconstructed graph nodes and reconstructed graph edges. The seventh generation unit is configured to obtain the trained model based on the training data set, the masking data set, the reconstructed graph nodes, and the reconstructed graph edges.

6. The apparatus of claim 5, wherein, The first extraction module comprises: The first extraction unit is configured to extract the semantic features based on the component feature information of each component. The second extraction unit is configured to extract the spatial features based on the position information of each component and the local spatial relationship between adjacent components. The third extraction unit is configured to extract the topological features based on the nesting relationship, connection relationship between different components, and the vertical relationship between the different components and the ground.

7. A device for graph structure representation of a BIM model and model error identification, characterized in that, The model training stage adopts the graph structure representation of the BIM model and the model error identification method according to any one of claims 1-2, and is applied to a model transfer learning stage, wherein the device comprises: The second extraction module is configured to extract semantic features, spatial features, and topological features between different types of components in the target BIM model based on component information of the different types of components. The generation module is configured to generate a target graph structure representation of the target BIM model based on the semantic features, spatial features, and topological features. The identification module is configured to input the target graph structure representation into a pre-trained transfer learning model to identify an error target identification result of the target BIM model, wherein the pre-trained transfer learning model is obtained by transfer learning of a trained training model.

8. The apparatus of claim 7, wherein, Further comprising: The third extraction module is configured to extract semantic features, spatial features, and topological features between different types of components in a BIM model containing an error identification target based on component information of the different types of components before inputting the target graph structure representation into the pre-trained transfer learning model. The second construction module is configured to generate an error graph structure representation of the BIM model containing the error identification target based on the semantic features, spatial features, and topological features, and to construct a transfer data set suitable for the pre-trained transfer learning model by using the error graph structure representation. The second training module is configured to train the pre-trained transfer learning model by using the transfer data set until a preset transfer condition is met to obtain a trained transfer learning model.

9. An electronic device, comprising: The memory, the processor, and the computer program stored on the memory and executable on the processor, wherein the processor executes the program to implement the graph structure representation of the BIM model and the model error identification method according to any one of claims 1-2 or the graph structure representation of the BIM model and the model error identification method according to any one of claims 3-4. ​ 10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The program is executed by a processor for implementing the graph structure representation of a BIM model and the model error identification method according to any one of claims 1-2 or the graph structure representation of a BIM model and the model error identification method according to any one of claims 3-4.

11. A computer program product, characterised in that, The computer program is executed for implementing the graph structure representation of a BIM model and the model error identification method according to any one of claims 1-2 or the graph structure representation of a BIM model and the model error identification method according to any one of claims 3-4.

Images