A node tracing and permission control method for BIM design changes of a large construction project

Abstract

The application relates to the field of BIM design, and discloses a node tracing and permission control method for BIM design changes of a large building project, which comprises the following steps: obtaining and analyzing a modification request data packet containing target change parameters to extract BIM component geometry data sets corresponding to the target change parameters; obtaining a node set in which structural connection correlations exist in a force topology diagram; constructing graph structure data in which three-dimensional vertex coordinates in the node set are taken as graph nodes and in which a correlation vector representing a force dependence relationship between adjacent building components is taken as a bearing relationship edge; extracting spatial interference features of the adjacent building components, and correcting attention weight parameters in a graph attention network model according to the spatial interference features; inputting the graph structure data into the graph attention network model configured with the updated attention weight, and outputting an access interception probability value for the modification request data packet; and executing a permission interception instruction when the access interception probability value is greater than an interception threshold value.

Description

Technical Field

[0001] This application relates to the field of BIM design, and in particular to a method for tracing nodes and controlling access permissions for BIM design changes in large-scale building projects. Background Technology

[0002] The BIM collaborative design method involves multiple design participants collaborating on the same building information model and updating its parameters via a network. In this method, any modification operation made by any participant to a model component must be reviewed and approved by the access control system, which determines whether to allow or block the change operation based on the structural relationships between the components.

[0003] In large-scale construction projects, steel structures and concrete structures are in a stress-dependent relationship through welded joints, anchor plates, and embedded connectors. When the geometric parameters of a building component are modified, the change in its stress distribution will be transmitted to adjacent components, causing spatial contour shifts in the adjacent components.

[0004] In related technologies, permission verification methods typically rely solely on user identification to search a pre-defined list of read and write permissions to determine whether to allow modification operations. The inventors discovered that this permission control method cannot determine the structural connection dependencies inherent in building components. When the parameters of a building component change, adjacent components with load-bearing dependencies may experience spatial interference and bounding box overlap due to the lack of load-bearing correlation verification, leading to inconsistencies in the 3D vertex coordinates in the building information model. Summary of the Invention

[0005] This application provides a method for node tracing and access control of BIM design changes in large-scale building projects, so as to at least partially solve the above-mentioned technical problems.

[0006] To achieve the above objectives, this application provides a method for node tracing and access control of BIM design changes in large-scale building projects, including: Obtain and parse the modification request data packet containing the target change parameters to extract the BIM component geometry dataset corresponding to the target change parameters; the BIM component geometry dataset contains the three-dimensional vertex coordinates and stress topology diagram of the corresponding building component in the building information model; Obtain the set of nodes with structural connections in the force topology diagram; Construct graph structure data with the coordinates of each three-dimensional vertex in the node set as graph nodes and the association vector representing the force dependence relationship between adjacent building components as bearing relationship edges; In response to structural association conflicts caused by the target change parameters, spatial interference features of adjacent building components are extracted, and the attention weight parameters in the graph attention network model are corrected according to the spatial interference features. The graph structure data is input into the graph attention network model configured with updated attention weights, and the access interception probability value for the modified request data packet is output. When the access interception probability value is greater than the interception threshold, the permission interception command is executed.

[0007] In this embodiment of the application, the above technical solution adds a verification step on the structural connection dependency of building components to the traditional control based solely on user identity and permissions. This ensures that parameter changes of force-dependent components are intercepted before they cause bounding box overlap and clipping, thus guaranteeing the data consistency of three-dimensional vertex coordinates in the building information model.

[0008] Other features and advantages of this application will be described in detail in the following detailed description section. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 This is a flowchart illustrating the steps of a method for tracing nodes and controlling access permissions for BIM design changes in a large-scale building project, provided in an exemplary embodiment of this application. Detailed Implementation

[0011] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the protection scope of this application.

[0012] This application provides a method for tracing and controlling access permissions for BIM design changes in large-scale building projects. Please refer to [link / reference]. Figure 1 The method for node tracing and access control of BIM design changes in large-scale building projects provided in this application includes the following steps: S101. Obtain and parse the modification request data packet containing the target change parameters to extract the BIM component geometry dataset corresponding to the target change parameters. Specifically, when a design participant in the BIM collaborative design platform initiates a modification request for building component parameters through a client, the modification request data packet is intercepted, and its protocol header field is parsed to read the target change parameters. The target change parameters include the unique identifier of the target component, the name of the geometric attribute to be modified, and the modified parameter value. Based on the target component identifier, the corresponding three-dimensional vertex coordinate dataset is indexed in the building information model database, and the structural force connection relationship record associated with the component is read to obtain the force topology diagram. The force topology diagram and the three-dimensional vertex coordinates together constitute the BIM component geometry dataset.

[0013] S102. Obtain the set of nodes with structural connections in the stress topology diagram. Specifically, traverse all connection relationship records in the stress topology diagram and extract the nodes corresponding to adjacent components that have direct structural connections with the target component as a node set; if the target component is a steel structure column and the stress topology diagram records concrete beam nodes and steel beam nodes connected to it, then include the beam nodes and steel beam nodes in the node set.

[0014] S103. Construct graph structure data with the coordinates of each 3D vertex in the node set as graph nodes and the association vector representing the force dependency relationship between adjacent building components as bearing relationship edges. Specifically, read the coordinates of the 3D vertex corresponding to each node in the node set, treat each 3D vertex coordinate as a graph node, read the material attribute field of each node in the node set, and obtain the corresponding initial material stiffness value by querying the stiffness coefficient table based on the material attribute field; calculate the Euclidean distance between the coordinates of the 3D vertices of two nodes with force dependency relationship, obtain the topological connection weight based on the initial material stiffness value and the Euclidean distance, and assign the topological connection weight to the association vector connecting the two nodes as the edge weight parameter of the bearing relationship edge, thus constructing graph structure data with graph nodes as vertices and bearing relationship edges as edges; this graph structure data simultaneously encodes the spatial geometric information and force connection relationship information of building components.

[0015] S104. In response to structural association conflicts caused by target parameter changes, spatial interference features of adjacent building components are extracted, and attention weight parameters in the graph attention network model are corrected based on the spatial interference features. Specifically, based on the target parameter changes and the 3D vertex coordinates, the coordinate deviation correlation matrix of the node set under the bearing relationship edge constraint is calculated. The partial derivative matrix of the coordinate deviation correlation matrix with respect to the target parameter changes is calculated to obtain the covariance gradient. When the absolute value of any element in the covariance gradient is greater than the conflict threshold, a structural association conflict is determined to exist. After determining that a conflict exists, a 3D bounding box corresponding to each building component is constructed for the 3D vertex coordinates belonging to different building components in the node set. The intersection volume value of the 3D bounding boxes of adjacent building components is calculated as a spatial interference feature. An interference penalty factor is calculated based on the intersection volume value, and the attention weight parameters in the graph attention network model are updated using the interference penalty factor.

[0016] S105. Input the graph structure data into the graph attention network model configured with updated attention weights, and output the access interception probability value for the modification request data packet. Specifically, the constructed graph structure data is sent to the forward computation channel of the graph attention network model; the model contains multiple graph attention layers and a multi-head attention computation structure. In each graph attention layer, each graph node calculates the aggregated node representation based on the feature information of its neighboring nodes and the current attention weight parameters. The multi-head attention computation structure concatenates the outputs of multiple attention heads and sends them to the fully connected layer for dimensional transformation, outputting a scalar value between 0 and 1 as the access interception probability value; this access interception probability value represents the degree to which the target change parameter poses a structural safety risk to adjacent components.

[0017] S106. When the access interception probability value is greater than the interception threshold, the permission interception command is executed to reject the writing of the modification request data packet; when the access interception probability value is not greater than the interception threshold, the modification request data packet is allowed to execute the writing of BIM component parameters. Specifically, the access interception probability value is compared with the preset interception threshold, which is preset according to the building structure safety level during the system initialization phase; if the access interception probability value is greater than the interception threshold, it indicates that the target change parameter will cause deformation of adjacent components beyond the allowable range of structural safety, the writing operation of the modification request data packet is rejected and the geometric data of the BIM component remains unchanged; if the access interception probability value is not greater than the interception threshold, it indicates that the structural association impact caused by the target change parameter is within an acceptable range, the modification request data packet is allowed, and the target change parameter is written into the parameter area of ​​the corresponding BIM component to complete the parameter update.

[0018] The above technical solution embeds the stress topology of building components into graph structure data. The graph attention network model is used to numerically identify structural association conflicts caused by parameter changes. When spatial interference is detected, the attention weight parameters are modified by the interference penalty factor to amplify the characteristic response of the conflicting components. Based on the traditional control that only relies on user identity and permissions, a verification link for the structural connection dependence of the building components themselves is added. This ensures that parameter changes of stress-dependent components are intercepted before bounding box overlap and clipping are triggered, thus guaranteeing the data consistency of three-dimensional vertex coordinates in the building information model.

[0019] In some embodiments, in response to structural association conflicts caused by the target change parameters, the method includes: S201, calculating the coordinate deviation association matrix of the node set under the bearing relationship edge constraint based on the target change parameters and the three-dimensional vertex coordinates. Specifically, the boundary point normal vectors of the surface mesh patches corresponding to the target change parameters are obtained, the dot product of the boundary point normal vectors and the bearing relationship edge direction vectors is calculated, and the normal deviation value is recorded when the dot product result is less than the collinearity threshold; the normal deviation value is injected as a disturbance variable into the spatial position distribution of the node set; the three-dimensional vertex coordinates of each node in the node set are arranged into an original coordinate matrix X according to the node number of the force topology diagram, and a mean-reduction operation is performed on each row of the original coordinate matrix X, that is, the mean vector of all node coordinates is subtracted from each node coordinate; the mean-reduction matrix is ​​multiplied by its transpose and divided by the total number of nodes minus one to obtain the spatial position covariance matrix; the coordinate deviation association matrix is ​​output based on the spatial position covariance matrix, where each element represents the degree of coordinate deviation association between different nodes under the bearing relationship edge constraint.

[0020] S202. Calculate the partial derivative matrix of the coordinate deviation correlation matrix with respect to the target change parameters to obtain the covariance gradient. Specifically, construct the Jacobian matrix of the coordinate deviation correlation matrix with respect to the target change parameters, that is, calculate the partial derivative of each element in the coordinate deviation correlation matrix for each component of the target change parameters to form a partial derivative matrix; traverse the positions of elements in the partial derivative matrix that have no connection with the bearing relationship edge, force the partial derivative values ​​of these positions to zero to perform matrix clipping, and output the covariance gradient based on the partial derivative matrix after matrix clipping.

[0021] S203. When the absolute value of any element in the covariance gradient is greater than the conflict threshold, a structural association conflict is determined to exist. Specifically, the absolute values ​​of all elements in the covariance gradient are traversed, and each absolute value is compared with the conflict threshold one by one. The conflict threshold is set in the initialization stage according to the structural safety level of the target building. Once the absolute value of any element is detected to be greater than the conflict threshold, it is determined that the current modification request will cause structural association deformation beyond the allowable range, confirming the existence of a structural association conflict and triggering the subsequent spatial interference feature extraction and attention weight correction process.

[0022] The above technical solution uses the partial derivative of the coordinate deviation correlation matrix to calculate the covariance gradient as a detection criterion for structural conflict. Compared with directly comparing the magnitude of node displacement, the covariance gradient reflects the direction and magnitude of the correlation change of displacement between nodes. It has a more sensitive detection capability for cooperative displacement under the constraint of bearing relationship, and avoids false interception when the overall structure has not yet become unstable due to the displacement of a single node exceeding the limit.

[0023] In some embodiments, spatial interference features of adjacent building components are extracted, and the attention weight parameters in the graph attention network model are corrected based on the spatial interference features, including: S301. Based on the coordinates of the 3D vertices belonging to different building components in the node set, construct the 3D bounding box corresponding to each building component. Specifically, for the target building component, extract all its corresponding 3D vertex coordinates to calculate the geometric center point. Subtract the geometric center point from each 3D vertex coordinate to obtain mean-centered coordinates. Construct a position covariance matrix based on the mean-centered coordinates. Perform eigenvalue decomposition on the position covariance matrix to obtain three orthogonal eigenvectors as the bounding box orientation axes. Project each 3D vertex coordinate onto the three orientation axes and take the difference between the maximum and minimum projection values ​​on each orientation axis as the bounding box side length, thus constructing a directed bounding box aligned with the actual orientation of the building component.

[0024] S302. Calculate the intersection volume of the three-dimensional bounding boxes of adjacent building components as a spatial interference feature. Specifically, extract the translation and rotation matrices corresponding to the target change parameters, and use the translation and rotation matrices to perform homogeneous coordinate transformations on the three-dimensional bounding boxes corresponding to each building component to generate a candidate bounding box group reflecting the changed spatial position. For any two adjacent three-dimensional bounding boxes in the candidate bounding box group, calculate the side length of the intersection region based on their minimum vertex coordinate set and maximum vertex coordinate set, and multiply the three-axis intersection side lengths to obtain the intersection volume value. This intersection volume value reflects the degree of spatial intrusion between adjacent components caused by the change of the target component parameters.

[0025] S303. Calculate the interference penalty factor based on the intersection volume value. Specifically, the intersection volume value is normalized by dividing it by the bounding box volume of the smaller of the two adjacent components, resulting in a normalized interference value between 0 and 1. This normalized interference value is then input into a preset penalty mapping function to calculate the interference penalty factor. The penalty mapping function ensures that the interference penalty factor gradually decreases from 1 as the intersection volume value increases from zero. The larger the intersection volume value, the smaller the penalty factor, and the stronger the suppression of the attention weight parameter.

[0026] S304. Update the attention weight parameters of the graph attention network model based on the interference penalty factor. Specifically, read the currently stored attention weight parameter matrix in the graph attention network model, multiply the interference penalty factor element by the weight value of the edge connected to the corresponding interference component in the attention weight parameter matrix, and write the product back to the model parameter area as the corrected attention weight parameters. After the update is completed, the graph attention network model will prioritize the adjacent component pairs that have spatial interference when performing forward calculation, increasing their feature contribution ratio to the access interception probability value.

[0027] The above technical solution constructs spatial interference features using the intersection volume of three-dimensional bounding boxes and corrects the attention weight parameters by using the interference penalty factor. This enables the graph attention network model to perceive the degree of spatial intrusion caused by parameter changes, assigning higher attention to components that cause severe interference, and ensuring the model's sensitivity to spatial interference scenarios.

[0028] In some embodiments, constructing graph structure data with the coordinates of each three-dimensional vertex in the node set as graph nodes and the association vector representing the force dependence relationship between adjacent building components as bearing relationship edges includes: S401. Read the material attribute fields from the stress topology diagram. Specifically, in addition to recording the connection relationships between building components, the stress topology diagram also stores a material attribute field for each component node. This field records the material type of the building component, such as steel structure, concrete structure, or steel-concrete composite structure, in the form of an enumeration value or a string. Read the material attribute fields of each node from the stress topology diagram and temporarily store them in the working memory.

[0029] S402. Obtain the corresponding initial stiffness value of the material by querying the stiffness coefficient table based on the material attribute field. Specifically, during the system initialization phase, the stiffness coefficients of commonly used building materials have been pre-set as a lookup table. This lookup table uses the material attribute field as the index key and the initial stiffness value of the material as the return value. The retrieved material attribute field is used as the index to query this lookup table to obtain the corresponding initial stiffness value of the material. The initial stiffness value of the material reflects the building material's ability to resist deformation in the elastic stage.

[0030] S403. Calculate the initial Euclidean distance between the coordinates of two 3D vertices connected by an associated vector. Specifically, for each connecting edge in the force topology diagram, read the 3D vertex coordinates of the two nodes it connects, call the arithmetic logic unit to calculate the square root of the sum of the squares of the coordinate differences between the two points in the three axes, and obtain the initial Euclidean distance; this initial Euclidean distance represents the spatial distance between the two building components when no parameter changes have occurred.

[0031] S404. Obtain the topological connection weight based on the initial material stiffness value and the initial Euclidean distance. Specifically, divide the initial material stiffness value by the square of the initial Euclidean distance to obtain the topological connection weight. The ratio of stiffness value to the square of distance is used as the weight because in structural mechanics, the mutual constraint force between components is directly proportional to stiffness and inversely proportional to the square of distance. This weight can accurately describe the transmission intensity of force influence between adjacent components.

[0032] S405. Assign the topological connection weights to the association vector as edge weight parameters for the bearing relationship edges. Specifically, write the calculated topological connection weights into the weight attribute field of the corresponding bearing relationship edge in the graph structure data; in the subsequent forward computation of the graph attention network model, these edge weight parameters will participate in the aggregation operation of node features together with the attention weight parameters, so that the stronger the force connection, the greater the influence of the component in the graph attention calculation.

[0033] Through the above technical solution, the material stiffness and spatial distance information of building components are encoded as edge weight parameters of the bearing relationship edges, so that the graph structure data records the connection relationship of components at the topological level and quantifies the transmission strength of the force influence at the mechanical level.

[0034] In some embodiments, based on the target change parameters and the three-dimensional vertex coordinates, the coordinate deviation correlation matrix of the node set under the bearing relationship edge constraint is calculated, including: S501. Obtain the boundary point normal vectors of the surface mesh patches corresponding to the target change parameters. Specifically, the geometric attribute name in the target change parameters points to a certain surface of the building component, such as the side of a column or the top surface of a beam; in the building information model database, index the surface mesh patch data corresponding to the geometric attribute name, and extract the normal vectors of each sampling point on the boundary of the patch as the boundary point normal vectors.

[0035] S502. Calculate the dot product of the boundary point normal vector and the bearing relationship edge direction vector. Specifically, for each bearing relationship edge, read the direction vector of the edge, which points from the starting node to the ending node connected by the edge; iterate through the boundary point normal vector and the bearing relationship edge direction vector one by one, and call the arithmetic logic unit to calculate the dot product of the two.

[0036] S503. When the dot product result is less than the collinearity threshold, record the normal deviation value. Specifically, if the absolute value of the dot product result is less than the preset collinearity threshold, it indicates that the boundary point normal vector and the bearing relationship edge direction vector are not collinear, that is, there is an angular deviation between the surface deformation direction caused by the target change parameters and the force transmission direction; calculate the angle between the normal vector and the edge direction vector as the normal deviation value and record it; the collinearity threshold is set in the initialization stage, and is usually taken as a value close to 1 to tolerate small directional deviations.

[0037] S504. Inject the normal deviation value as a perturbation variable into the spatial location distribution of the node set. Specifically, the recorded normal deviation value is decomposed into three-axis perturbation components according to the direction vector of the bearing relationship edge, and then superimposed onto the three-dimensional vertex coordinates of the corresponding node to generate spatial location distribution data containing perturbation variables.

[0038] S505. The coordinate deviation correlation matrix is ​​calculated based on the spatial location distribution of the injected perturbation variables. Specifically, the perturbation three-dimensional coordinates of each graph node with perturbation variables are extracted from the spatial location distribution. The mean vector of the perturbation three-dimensional coordinates of all graph nodes is calculated. The vector difference between the perturbation three-dimensional coordinates of each graph node and the mean vector is calculated to obtain the corresponding spatial deviation vector. All spatial deviation vectors are concatenated according to the node index of the force topology diagram to form a global deviation matrix. The global deviation matrix is ​​multiplied by its transpose to obtain the coordinate deviation correlation matrix.

[0039] The above technical solution introduces the normal deviation between the surface deformation direction and the force transmission direction caused by the target parameter change as a disturbance variable into the calculation of the coordinate deviation correlation matrix. This allows the detection of structural correlation conflicts to consider both the magnitude of the nodal displacement and the degree of matching between the displacement direction and the force direction, thus filtering out invalid displacements in non-force directions.

[0040] In some embodiments, the intersection volume of the three-dimensional bounding boxes of adjacent building components is calculated as the spatial interference feature; an interference penalty factor is calculated based on the intersection volume value, including: S601. Extract the translation and rotation matrices corresponding to the target change parameters. Specifically, the target change parameters include the amount of modification to the geometric parameters of the building components. This amount of modification can be expressed as the product of a fourth-order translation matrix and a fourth-order rotation matrix through homogeneous coordinate transformation. Read the amount of modification from the modification request data packet and construct the transformation matrix.

[0041] S602. Using translation and rotation matrices, perform homogeneous coordinate transformations on the 3D bounding boxes corresponding to each building component to generate candidate bounding box groups. Specifically, convert the vertex coordinates of the 3D bounding boxes of each building component into homogeneous coordinate form, and then multiply them by the rotation and translation matrices on the left to obtain the transformed vertex coordinate set. Take the minimum and maximum values ​​of the transformed vertex coordinate set in each axis to reconstruct the bounding box, thus obtaining a candidate bounding box group that reflects the spatial position after the target change parameters are executed.

[0042] S603. For any two adjacent 3D bounding boxes in the candidate bounding box group, calculate the intersection volume based on their minimum and maximum vertex coordinate sets. Specifically, for each pair of adjacent bounding boxes in the candidate bounding box group, take the smaller value of each axis component of the maximum vertex coordinates of the two bounding boxes to form the maximum point of the intersection region, and take the larger value of each axis component of the minimum vertex coordinates of the two bounding boxes to form the minimum point of the intersection region; if the minimum point in each of the three axes is less than the corresponding maximum point, then the intersection region exists, and the intersection volume is the product of the lengths of the intersection intervals of the three axes; if there is no intersection interval in any axis direction, then the intersection volume value is zero.

[0043] S604. Normalize the intersecting volume value by dividing it by the volume of the smaller of the two adjacent bounding boxes to obtain the normalized interference value. Specifically, calculate the volume value of each of the two bounding boxes, which is the product of the three-axis side lengths, take the smaller one as the normalization denominator, and divide the intersecting volume value by this denominator to obtain the normalized interference value in the range of 0 to 1.

[0044] S605. The normalized interference value is input into a preset penalty mapping function to calculate the interference penalty factor. Specifically, the penalty mapping function is set during the initialization phase. When the normalized interference value is 0, the penalty factor is 1, meaning no penalty is applied. As the normalized interference value increases, the penalty factor monotonically decreases. When the normalized interference value approaches 1, the penalty factor approaches 0, meaning that the corresponding attention weight parameter is almost suppressed in the graph attention calculation. This penalty mapping function can be a Sigmoid-like function or a piecewise linear function.

[0045] The above technical solution uses the normalized interference value as the input of the penalty mapping function to calculate the interference penalty factor, so that the intensity of the interference penalty corresponds to the proportion of the degree of intersection of the bounding boxes. When there is slight interference, the correction amplitude of the attention weight parameter is small to preserve the transmission of information related to the normal structure, while when there is severe interference, it is suppressed to highlight the conflict features.

[0046] In some embodiments, the coordinate deviation correlation matrix is ​​calculated based on the spatial location distribution of the injected perturbation variables, including: S701. Extract the perturbation 3D coordinates of each graph node with perturbation variables from the spatial location distribution. Specifically, traverse the node set and read the 3D coordinate values ​​of each graph node after the injection of normal deviation perturbation. The perturbation 3D coordinates are obtained by superimposing the original coordinates and the perturbation components, representing the possible spatial displacement of adjacent components after the target change parameters are applied.

[0047] S702. Calculate the mean vector of the perturbed 3D coordinates of all graph nodes. Specifically, sum the perturbed 3D coordinates of all graph nodes along the three axes, and divide the sum of each axis by the total number of graph nodes to obtain a three-element mean vector; this mean vector represents the geometric center position of the node set after perturbing.

[0048] S703. Calculate the vector difference between the perturbation 3D coordinates and the mean vector of each graph node to obtain the corresponding spatial deviation vector. Specifically, subtract the mean vector from the perturbation 3D coordinates of each node to obtain the spatial deviation vector of each graph node; the spatial deviation vector represents the spatial offset of each node relative to the geometric center, eliminating the influence of overall translation and retaining only the relative displacement information between nodes.

[0049] S704. Concatenate all spatial deviation vectors according to the node numbers of the force topology diagram to form a global deviation matrix. Specifically, according to the numbering order of each node in the force topology diagram, arrange the corresponding spatial deviation vectors sequentially into rows of the global deviation matrix. The number of rows in this matrix is ​​the total number of nodes, and the number of columns is the spatial dimension, i.e., three columns.

[0050] S705. Multiply the global deviation matrix by its transpose to obtain the coordinate deviation correlation matrix. Specifically, call the matrix multiplication operation unit to multiply the global deviation matrix by its transpose to obtain a square matrix whose number of rows and columns are both the total number of nodes, i.e., the coordinate deviation correlation matrix; the element in the i-th row and j-th column of this matrix represents the inner product between the spatial deviation vectors of the i-th node and the j-th node, reflecting the correlation strength between the displacement changes of the two nodes.

[0051] The above technical solution constructs a global deviation matrix using the mean-centered spatial deviation vector and calculates the autocorrelation to obtain the coordinate deviation correlation matrix. This eliminates the overall translational degrees of freedom of the node set. The matrix captures the correlation pattern of relative displacement between nodes, so that the subsequent covariance gradient calculation is not affected by the choice of coordinate system.

[0052] In some embodiments, calculating the partial derivative matrix of the coordinate deviation correlation matrix with respect to the target change parameter to obtain the covariance gradient includes: S801. Construct the Jacobian matrix of the coordinate deviation correlation matrix with respect to the target change parameters. Specifically, the target change parameters include multiple scalar components such as displacement, rotation angle, and size parameter values; calculate the partial derivative of each element in the coordinate deviation correlation matrix with respect to each component, and organize all partial derivatives into a Jacobian matrix according to the independent variable components and the matrix element positions; the rows of this Jacobian matrix correspond to the element positions of the coordinate deviation correlation matrix, and the columns correspond to each component of the target change parameters.

[0053] S802. Traverse the Jacobian matrix and force the corresponding terms of nodes without load-bearing relationships to zero to perform matrix pruning. Specifically, read the adjacency matrix of the force topology graph, traverse the row and column indices corresponding to each element in the Jacobian matrix, and if the node pair corresponding to the index does not have a load-bearing relationship in the force topology graph, force the partial derivative value at that position in the Jacobian matrix to zero. After pruning, the Jacobian matrix only retains the partial derivative terms corresponding to node pairs with force-bearing relationships, eliminating the noise of partial derivatives for node pairs without force-bearing relationships.

[0054] S803. Output the covariance gradient based on the Jacobian matrix after matrix clipping. Specifically, multiply the clipped Jacobian matrix by its transpose and take the trace of the result. The resulting scalar value is the covariance gradient; this change reflects the total magnitude of the displacement correlation change of nodes with force connection relationships under the action of target changing parameters.

[0055] By employing the above technical solution, partial derivative terms of edges without load-bearing relationships are pruned when calculating the covariance gradient, ensuring that the change only reflects the displacement changes of node pairs with force-dependent relationships. This approach is insensitive to parameter changes of force-independent nodes, thus avoiding false alarms for non-structurally related regions.

[0056] In some embodiments, based on the coordinates of the three-dimensional vertices belonging to different building components in the node set, a three-dimensional bounding box corresponding to each building component is constructed, including: S901. For the target building component, extract the coordinates of all its corresponding three-dimensional vertices and calculate the geometric center point. Specifically, obtain the coordinates of all vertices of the surface mesh of the component from the building information model database based on the component's unique identifier. Sum the coordinates of each axis coordinate component and divide by the total number of vertices to obtain the coordinates of the geometric center point of the component.

[0057] S902. Calculate the mean-centered coordinates of each 3D vertex relative to the geometric center point. Specifically, subtract the corresponding component of the geometric center point from the 3D coordinates of each vertex to obtain the mean-centered coordinates; after centering, the mean vector of the vertex coordinates is the zero vector, and the origin of the coordinate system is translated to the geometric center of the component.

[0058] S903. Construct a position covariance matrix based on mean-centered coordinates. Specifically, arrange the mean-centered coordinates of all vertices into an N-row, 3-column matrix. Multiply the transpose of this matrix by itself and divide by the total number of vertices N to obtain a 3-row, 3-column position covariance matrix. The diagonal elements of this matrix represent the variance of the vertex along the three coordinate axes, and the off-diagonal elements represent the covariance between different coordinate axes.

[0059] S904. Perform eigenvalue decomposition on the position covariance matrix to obtain three orthogonal eigenvectors as the bounding box orientation axes. Specifically, the eigenvalue decomposition operation unit is called to decompose the position covariance matrix to obtain three eigenvalues ​​and three corresponding orthogonal eigenvectors; the three eigenvectors are sorted in descending order of their corresponding eigenvalues ​​and serve as the first, second, and third orientation axes of the bounding box, respectively; these three orientation axes are aligned with the main directions of the component vertex distribution, so that the bounding box wraps the component as tightly as possible.

[0060] S905. Project the coordinates of each 3D vertex onto the three direction axes, and take the difference between the maximum and minimum projection values ​​on each direction axis as the bounding box side length. Specifically, take the dot product of the original 3D coordinates of each vertex with the three direction axes to obtain the projection value, count the maximum and minimum projection values ​​on each direction axis, and the difference between the maximum and minimum values ​​is the bounding box side length on that direction axis. Determine the size of the bounding box by obtaining the side length values ​​on the three direction axes.

[0061] The above technical solution uses the eigenvalue decomposition of the position covariance matrix to determine the orientation axis of the bounding box, so that the bounding box is aligned with the actual spatial distribution direction of the component vertices. Compared with the bounding box aligned with the axis, it more tightly wraps the inclined or irregular building components, reduces the invalid gap between the bounding boxes, and improves the representativeness of the intersection volume value to the real spatial interference.

[0062] In some embodiments, executing permission interception instructions includes: S1001. Obtain the starting address of the computer memory page corresponding to the BIM component geometric dataset. Specifically, when loading the building information model, the geometric data of each component is loaded into the computer memory and the allocated starting address of the memory is recorded; when an access control command is triggered, the memory mapping table is queried based on the unique identifier of the target component to obtain the starting address of the memory page currently occupied by its BIM component geometric dataset.

[0063] S1002. Obtain the offset range of the BIM component geometric dataset in computer memory. Specifically, query the dataset byte length recorded in the memory mapping table, take the memory start address as the base address, and add the start address to the byte length as the end address to obtain the offset range; this offset range covers the memory occupied area of ​​all geometric data of the target component.

[0064] S1003. A read-only identifier containing the memory start address and offset range is sent to the memory management unit. The memory management unit then marks the corresponding memory page in the page table as write-protected, thereby blocking the write queue containing the memory start address. Specifically, the memory start address and offset range are encapsulated into a read-only identifier data packet and sent to the operating system's memory management unit via a system call interface. After receiving the identifier, the memory management unit locates the corresponding memory page entry in the page table and sets the write protection flag of that entry to a valid state. Subsequently, any write instructions targeting that memory region will be intercepted by the hardware access control of the memory management unit. The write operation queue for the target component parameters in the BIM software cannot be executed, thus achieving low-level permission interception.

[0065] The above technical solution decentralizes the implementation of permission interception to the page table write protection level of the memory management unit. By using hardware memory access control, it blocks the writing of data memory areas of the target component. Compared with intercepting API calls at the application layer, this method is not affected by changes in BIM software version or interface, and the interception is more reliable and cannot be bypassed by the application layer.

[0066] In some embodiments, the memory management unit marks the corresponding memory page as write-protected in the page table to block the write queue containing the memory start address, including: S1101. Construct an asynchronous interception queue targeting the memory start address. Specifically, allocate a fixed-length queue buffer in memory. This queue is used to temporarily store write protection instructions to be executed. Each queue element records the memory start address, offset range, and control instruction type.

[0067] S1102. Push the write-protected memory control instructions into the asynchronous intercept queue. Specifically, package the write-protected instructions containing the target memory page address range into a control instruction structure, append it to the tail of the asynchronous intercept queue, and update the tail pointer.

[0068] S1103. Monitor the real-time rendering timestamps of non-conflicting components in the underlying graphics rendering pipeline of the current building information modeling software. Specifically, continuously read the real-time rendering timestamps of the rendering pipeline through the performance counter interface of the graphics rendering pipeline. These timestamps record the progress of the current rendering frame. Here, the rendering timestamps of non-conflicting components, i.e., other components that are not affected by parameter changes, are monitored to determine whether the rendering pipeline is in an idle period.

[0069] S1104. When the real-time rendering timestamp indicates that the current frame is in the gap period between adjacent rendering frames, the memory control instructions in the asynchronous intercept queue are triggered to perform a memory page marking operation. Specifically, when the real-time rendering timestamp indicates that the current rendering frame has been submitted to the display buffer and the drawing instructions for the next frame have not yet arrived, it is determined that the frame is in the gap period between adjacent rendering frames. The control instructions in the asynchronous intercept queue are triggered to instruct the memory management unit to perform a page table write protection marking operation. Page table marking is performed during the rendering gap to avoid bus conflicts between page table modification and rendering pipeline memory access.

[0070] By utilizing the above technical solution, page table write protection operations are performed periodically during the rendering frame intervals. This aligns the timing of memory control commands with the idle window of the rendering pipeline, avoiding bus access conflicts or rendering anomalies caused by modifying page table permissions while the rendering pipeline is reading target component data. This ensures that the interception operation is imperceptible to the normal operation of the BIM software.

[0071] In some embodiments, the method further includes a dimension alignment operation for the graph attention network model, including: S1201, obtaining the baseline tensor dimension features of the graph attention network model and extracting the current tensor dimension features of the covariance gradient. Specifically, the graph attention network model fixes the dimension parameters of the input layer, i.e., the baseline tensor dimension features, during initialization; after calculating the covariance gradient, its row and column numbers are read as the current tensor dimension features.

[0072] S1202. When the current tensor dimension is smaller than the reference tensor dimension, perform zero-padded operation on the covariance gradient to align it with the reference tensor dimension and generate the target covariance gradient. Specifically, add rows and columns of all zeros to the right and below the covariance gradient matrix to make the dimension of the padded matrix equal to that of the reference tensor dimension. The padded process does not change the position or value of the original elements in the covariance gradient. The padded matrix is ​​then fed into the graph attention network model as the target covariance gradient.

[0073] S1203. When the current tensor dimension is greater than the reference tensor dimension, perform max pooling dimensionality reduction on the covariance gradient to align it with the reference tensor dimension and generate the target covariance gradient. Specifically, the covariance gradient is divided into a sliding window grid corresponding to the reference tensor dimension, and the maximum value of each element in the window is taken as the output element, reducing the dimension of the covariance gradient to be consistent with the reference tensor dimension. The step size and window size of the pooling window are determined based on the ratio of the current dimension to the reference dimension.

[0074] By using the above technical solution, a dimension alignment operation is performed before the covariance gradient is fed into the graph attention network model, so that the covariance gradients calculated under different node sizes are uniformly input into the network model with a fixed input dimension, eliminating the need to train models with different input dimensions for building projects of different sizes.

[0075] In some embodiments, before inputting the graph structure data into the graph attention network model configured with updated attention weights and outputting the access interception probability value for the modification request data packet, the method further includes: S1301, obtaining the surface mesh patch corresponding to the target change parameter, and calculating the surface principal curvature value corresponding to each graph node in the surface mesh patch. Specifically, the mesh patch data of the target component change surface is read from the building information model database, and the maximum principal curvature value and minimum principal curvature value of each vertex on the patch are calculated using a surface fitting method. The average of the two is taken as the surface principal curvature value of the graph node; the surface principal curvature value reflects the degree of curvature of the component surface at that vertex.

[0076] S1302. Calculate the absolute difference of the principal curvature values ​​of the surfaces between two adjacent graph nodes to obtain the curvature change rate parameter. Specifically, traverse the two graph nodes connected by each load-bearing relationship edge in the graph structure data, and calculate the absolute difference of the principal curvature values ​​of the surfaces of the two nodes as the curvature change rate parameter; the larger the curvature change rate parameter, the more significant the difference in surface geometric features between the two adjacent components at the connection, and the higher the degree of stress concentration caused by the change may be.

[0077] S1303. Generate a curvature correction matrix based on the curvature change rate parameter. Specifically, construct a correction matrix of the same size as the attention weight parameter matrix, and use the curvature change rate parameter corresponding to each connecting edge as the value of the corresponding diagonal element in the matrix, while setting the off-diagonal elements to 1; the curvature change rate parameter is mapped to a correction multiplier greater than or equal to 1 through a preset activation function.

[0078] S1304. Multiply the curvature correction matrix element by element by the attention weight parameters to obtain the curvature-corrected attention weight parameters. Specifically, call the matrix element-wise multiplication operation unit to multiply the curvature correction matrix by the corresponding element values ​​of the current attention weight parameter matrix. The product is the curvature-corrected attention weight parameters. The greater the curvature difference of the component connection edge, the higher the amplification factor of its attention weight parameters, and the higher the attention of the graph attention network model to the connection edge.

[0079] Through the above technical solution, a curvature correction matrix is ​​further introduced on the basis of correcting the attention weight parameter by the interference penalty factor, so that the connection edge of the component with large differences in surface geometric features receives higher attention, which makes up for the lack of attention to the hidden structural conflicts that are not penetrating but have a high risk of stress concentration based solely on volume interference judgment.

[0080] In some embodiments, before inputting the graph structure data into the graph attention network model configured with updated attention weights and outputting the access interception probability value for the modification request data packet, the method further includes: S1401, calculating the coordinate change magnitude corresponding to each graph node based on the target change parameters. Specifically, the displacement, rotation, and size change included in the target change parameters are applied to the three-dimensional vertex coordinates of each graph node, and the Euclidean distance between the coordinates before and after the change is calculated as the coordinate change magnitude; this magnitude represents the degree to which the node is affected by the target change parameters.

[0081] S1402. Perform filtering operator calculations on the coordinates of each 3D vertex in the graph structure data. Specifically, call the mean filter or median filter operator to smooth the coordinates of each 3D vertex. The filtering operator eliminates random measurement errors introduced during the modeling process and local coordinate jitter caused by mesh generation.

[0082] S1403. Filter out nodes with small-amplitude deformation whose coordinate change range is less than the modeling error tolerance, and retain nodes with large-amplitude deformation whose coordinate change range is greater than or equal to the modeling error tolerance. Specifically, compare the coordinate change range of each node with the modeling error tolerance, which is set in the initialization stage according to the design accuracy level of the BIM model and is taken as a proportion of the model design tolerance; mark nodes with a change range less than the tolerance as small-amplitude deformation nodes and filter them out, and retain nodes with a change range greater than or equal to the tolerance for large-amplitude deformation.

[0083] S1404. Input only the graph nodes corresponding to the nodes with large deformations into the graph attention network model. Specifically, remove the nodes with small deformations and their connected bearing relationship edges from the graph structure data, and retain only the nodes with large deformations and their adjacent edges to form the reduced graph structure data. Then, feed the reduced graph structure data into the graph attention network model for forward computation.

[0084] By using the above technical solution, nodes whose coordinate changes do not exceed the tolerance of modeling error are filtered out before model inference. This avoids unnecessary attention weight correction and model calculation triggered by small coordinate offsets within the modeling tolerance range, reduces the computation node scale of the graph attention network model, and improves the response speed of permission determination.

[0085] In some embodiments, inputting the graph structure data into the graph attention network model configured with updated attention weights further includes: S1501. Input the graph structure data into a graph attention network model containing multiple graph attention layers and a multi-head attention calculation method to obtain a high-dimensional spatial association representation of each graph node. Specifically, the graph attention network model consists of K stacked graph attention layers, each layer configured with H independent attention heads. In each graph attention layer, each attention head independently calculates the attention score between adjacent graph nodes, and the features of neighboring nodes are weighted and aggregated according to the attention score and then concatenated with the features of the current node. The outputs of the H attention heads are linearly transformed and nonlinearly activated as the output of that layer. The multi-layer stacking allows the representation of each graph node to aggregate the structural information of multiple neighbors, outputting a high-dimensional spatial association representation of each graph node. This high-dimensional spatial association representation simultaneously encodes the geometric features, force connection relationships, and spatial interference degree of the nodes.

[0086] The above technical solution uses a multi-layer graph attention layer and a multi-head attention calculation method to capture the association patterns between graph nodes from different subspaces. The multi-layer stacking enables the node representation to aggregate the indirect force transmission information of distant associated components. The multi-head calculation method avoids the overfitting of a single attention head to a specific association pattern and improves the accuracy of the access interception probability value in the discrimination of complex structural association scenarios.

[0087] In some embodiments, after outputting the access interception probability value for the modification request data packet, the method further includes: S1601, when the access interception probability value is not greater than the interception threshold but greater than the warning threshold, extracting the material attribute fields corresponding to adjacent building components in the stress topology diagram, and using them as the first structural material identifier and the second structural material identifier, respectively. Specifically, the warning threshold is a preset proportion value of the interception threshold, such as 70%, set during system initialization; when the access interception probability value falls into the gray area between the warning threshold and the interception threshold, reading the material attribute fields of the components adjacent to the target component in the stress topology diagram, and recording them as the first structural material identifier and the second structural material identifier, respectively.

[0088] S1602. When the first structural material identifier represents a steel structure and the second structural material identifier represents a concrete structure, the interception threshold is reduced to obtain the re-estimation threshold. Specifically, the two material identifiers are compared. When they represent a steel structure and a concrete structure, respectively, the interception threshold is multiplied by a material adjustment coefficient less than 1 to obtain the re-estimation threshold. This material adjustment coefficient is set during initialization based on the failure test data of the steel-concrete connection node, reflecting that the steel-concrete dissimilar material connection has a higher structural safety risk when parameters change, requiring a stricter interception threshold.

[0089] S1603. When the access interception probability value is greater than the re-evaluation threshold, execute the permission interception command. Specifically, compare the original access interception probability value with the reduced re-evaluation threshold again. If the probability value is greater than the re-evaluation threshold, execute the permission interception command; if the probability value is still not greater than the re-evaluation threshold, allow the modification request data packet to pass.

[0090] The above technical solution introduces a material-sensitive reassessment strategy when the access interception probability value is in the gray range. This reduces the interception threshold for the high-risk structural safety scenario of steel-concrete dissimilar material connection, avoids missed detection of dissimilar material connection scenarios due to a uniform threshold, and does not affect the normal threshold determination of same material connection scenarios.

[0091] In some embodiments, the graph structure data is input into the graph attention network model configured with updated attention weights, and the access interception probability value for the modified request data packet is output, including: S1701. The output features of the last graph attention layer are concatenated with the output features of all previous graph attention layers to generate a multi-layer fusion feature. This multi-layer fusion feature is then fed into a fully connected layer, which outputs the access interception probability value. Specifically, each graph attention layer in the graph attention network model outputs an intermediate feature representation for each graph node during the forward computation. After the last layer is computed, the graph node features output from each layer are read, and the features from the first to the Kth layers are concatenated sequentially along the node dimension to form a multi-layer fusion feature. This multi-layer fusion feature is then fed into a fully connected layer, which performs a linear transformation on the fusion feature and outputs an access interception probability value in the range of 0 to 1 using the Sigmoid activation function. The concatenation of multi-layer features allows the output to comprehensively utilize structural association information at different levels of abstraction. Shallow features retain local geometric details, while deep features encode the global force propagation path.

[0092] By using the above technical solution, the output features of each graph attention layer are spliced ​​into multi-layer fusion features and then sent to the fully connected layer to output the access interception probability value. This enables the final judgment to comprehensively utilize the local geometric features of the shallow layer and the global force propagation features of the deep layer, avoiding the loss of shallow structural details due to using only the last layer features, and improving the judgment stability under different structural complexity scenarios.

[0093] All of the above-mentioned optional technical solutions can be combined in any way to form the optional embodiments of this application, and will not be described in detail here.

[0094] The above are merely preferred embodiments of this application and are not intended to limit this application in any way. Any simple modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of this application without departing from the scope of the technical solution of this application shall still fall within the scope of the technical solution of this application.

Claims

1. A method for node tracing and access control of BIM design changes in large-scale building projects, characterized in that, include: Obtain and parse the modification request data packet containing the target change parameters to extract the BIM component geometry dataset corresponding to the target change parameters; The BIM component geometry dataset contains the three-dimensional vertex coordinates and stress topology diagram of the corresponding building component in the building information model; Obtain the set of nodes with structural connections in the force topology diagram; Construct graph structure data with the coordinates of each three-dimensional vertex in the node set as graph nodes and the association vector representing the force dependence relationship between adjacent building components as bearing relationship edges; In response to structural association conflicts caused by the target change parameters, spatial interference features of adjacent building components are extracted, and the attention weight parameters in the graph attention network model are corrected according to the spatial interference features. The graph structure data is input into the graph attention network model configured with updated attention weights, and the access interception probability value for the modified request data packet is output. When the access interception probability value is greater than the interception threshold, the permission interception command is executed.

2. The method according to claim 1, characterized in that, In response to structural association conflicts caused by the target change parameters, including: Based on the target change parameters and the three-dimensional vertex coordinates, calculate the coordinate deviation correlation matrix of the node set under the bearing relationship edge constraint; Calculate the partial derivative matrix of the coordinate deviation correlation matrix with respect to the target change parameter to obtain the covariance gradient; When the absolute value of any element within the covariance gradient is greater than the conflict threshold, it is determined that the structural association conflict exists.

3. The method according to claim 2, characterized in that, Extracting spatial interference features of adjacent building components, and correcting the attention weight parameters in the graph attention network model based on the spatial interference features, including: Based on the coordinates of the three-dimensional vertices belonging to different building components in the node set, construct the three-dimensional bounding box corresponding to each building component; The intersection volume of the three-dimensional bounding boxes of adjacent building components is calculated as the spatial interference feature; an interference penalty factor is calculated based on the intersection volume value. The attention weight parameters of the graph attention network model are updated based on the interference penalty factor; wherein the initial weight matrix is ​​used as the baseline attention weight of the graph attention network model; and the interference penalty factor is combined with the baseline attention weight to update its weight parameters.

4. The method according to claim 3, characterized in that, Constructing graph structure data with the coordinates of each three-dimensional vertex in the node set as graph nodes and the association vector representing the force dependence relationship between adjacent building components as bearing relationship edges, including: Read the material attribute fields from the force topology diagram; Based on the material attribute fields, query the stiffness coefficient table to extract the corresponding initial stiffness value of the material; Calculate the initial Euclidean distance between the coordinates of the two three-dimensional vertices connected by the associated vector; The topology connection weight is obtained based on the initial stiffness value of the material and the initial Euclidean distance; The topological connection weights are configured as edge attribute features corresponding to the bearing relationship edges to generate the graph structure data.

5. The method according to claim 4, characterized in that, Based on the target change parameters and the three-dimensional vertex coordinates, calculate the coordinate deviation correlation matrix of the node set under the bearing relationship edge constraint, including: Obtain the boundary point normal vector of the surface mesh patch corresponding to the target change parameter; Calculate the dot product of the boundary point normal vector and the bearing relationship edge direction vector; When the dot product result is less than the collinearity threshold, the normal deviation value is recorded; The normal deviation value is injected as a perturbation variable into the spatial location distribution of the node set; The coordinate deviation correlation matrix is ​​calculated based on the spatial location distribution of the injected perturbation variables.

6. The method according to claim 5, characterized in that, The intersection volume of the three-dimensional bounding boxes of adjacent building components is calculated as the spatial interference feature; The interference penalty factor is calculated based on the intersecting volume value, including: Extract the model translation and rotation matrices corresponding to the target change parameters; Using the model translation matrix and the rotation matrix, homogeneous coordinate transformations are performed on the three-dimensional bounding boxes corresponding to each building component to generate candidate bounding box groups; For any two adjacent 3D bounding boxes in the candidate bounding box group, the intersecting cube is obtained based on their minimum and maximum vertex coordinate sets; The product of the axial components of the intersecting cubes is calculated to obtain the intersecting volume value; Calculate the ratio of the intersecting volume value to the sum of the total volumes of the two adjacent 3D bounding boxes to generate the interference ratio parameter; The interference penalty factor is obtained based on the exponential calculation result of the interference ratio parameter and the attenuation reference value.

7. The method according to claim 6, characterized in that, The coordinate deviation correlation matrix is ​​calculated based on the spatial location distribution of the injected perturbation variables, including: Extract the perturbation three-dimensional coordinates of each graph node with the perturbation variable from the spatial location distribution; Calculate the mean vector of the perturbation three-dimensional coordinates of all the graph nodes; Calculate the vector difference between the perturbation three-dimensional coordinates of each graph node and the mean vector to obtain the corresponding spatial deviation vector; All the spatial deviation vectors are concatenated into a global deviation matrix according to the node numbers of the force topology diagram; Calculate the product matrix of the global deviation matrix and its transpose; The coordinate deviation correlation matrix is ​​generated based on the ratio of the product matrix to the total number of nodes in the node set.

8. The method according to claim 7, characterized in that, Calculate the partial derivative matrix of the coordinate deviation correlation matrix with respect to the target change parameter to obtain the covariance gradient, including: Construct the Jacobian matrix of the coordinate deviation correlation matrix with respect to the target change parameters; Traverse the Jacobian matrix and force the corresponding items connected by edges without the aforementioned carrying relationship to zero to perform matrix pruning; The covariance gradient is output based on the Jacobian matrix after matrix clipping.

9. The method according to claim 8, characterized in that, Based on the coordinates of the three-dimensional vertices belonging to different building components in the node set, a three-dimensional bounding box corresponding to each building component is constructed, including: For the target building component, extract the coordinates of all its corresponding three-dimensional vertices and calculate the geometric center point; Calculate the mean-centered coordinates of each of the three-dimensional vertex coordinates relative to the geometric center point; Construct a position covariance matrix based on the mean-centered coordinates; The position covariance matrix is ​​decomposed into eigenvalues ​​to obtain three orthogonal eigenvectors as the bounding box orientation axes; Project the coordinates of all three-dimensional vertices of the target building component onto the bounding box direction axis, and extract the maximum and minimum projection values ​​in each direction; Based on the bounding box orientation axis and the corresponding maximum and minimum projection values, a directed 3D bounding box is constructed and generated as the 3D bounding box.

10. The method according to claim 9, characterized in that, Execution permission interception commands, including: Obtain the starting address of the computer memory page corresponding to the BIM component geometry dataset; Obtain the offset range of the BIM component geometry dataset in computer memory; A read-only identifier containing the memory start address and the offset range is sent to the memory management unit; The memory management unit marks the corresponding memory page in the page table as write-protected to block the write queue containing the memory start address.

Images