A project auxiliary division method based on BIM technology

Abstract

The application provides a project auxiliary division method based on BIM technology, which comprises the following steps: extracting geometric information, system information and construction information of BIM components, constructing a multi-dimensional feature vector containing normalized spatial coordinates, system topology coding and construction process sequence, and generating a component comprehensive connection strength matrix by weighted fusion of spatial distance, topology correlation degree and construction proximity. Based on the matrix, an anisotropic potential field base function is constructed and a scalar potential field is generated by superposition, a local minimum value of the inverse scalar potential field is solved as a clustering center, and an initial division boundary is obtained through a three-dimensional watershed algorithm. Then, according to the key system path integrity constraint, if there is a logical break, the break sample is extracted and the weight parameter is adjusted through a heuristic optimization algorithm, the division result is iteratively updated until the path integrity requirement is met, and the project work set division scheme is output.

Description

Technical Field

[0001] This application pertains to the field of project partitioning, and in particular relates to a project partitioning method based on BIM technology. Background Technology

[0002] With the widespread application of Building Information Modeling (BIM) technology throughout the entire lifecycle of building projects, it has become an important digital carrier for multi-disciplinary collaborative design, construction process management, and post-construction operation and maintenance. Modern building projects are constantly expanding in scale, and structural and electromechanical systems are becoming increasingly complex. A complete BIM model typically contains a large number of architectural, structural, and electromechanical components. To meet the needs of parallel collaboration among multiple teams in large-scale projects, rational allocation of hardware and software resources, and refined management of construction procedures, it is necessary to rationally divide the large and complex global BIM model into several sub-work sets or project partitions. Since BIM components not only have a clear three-dimensional spatial distribution but also possess system connection topology relationships and are constrained by the order of construction procedures during actual construction, project partitioning cannot be based solely on a single spatial dimension.

[0003] In current BIM project management practices, the partitioning of model components into work sets typically relies on the experience of engineers or is based solely on single rules such as floor elevation, spatial grid, or professional category. This makes it difficult to comprehensively consider the spatial proximity, system topology, and construction sequence relationships between components. Such partitioning methods can easily sever component sets that should be collaboratively processed, potentially creating unreasonable logical breaks in critical system paths such as MEP (Mechanical, Electrical, and Plumbing) systems and load-bearing structures. This increases the management costs of subsequent collaborative modifications, information verification, and construction organization, and may even lead to construction conflicts and rework. Therefore, how to combine the multi-dimensional relational information of BIM components to achieve a project-aided partitioning that balances spatial rationality, system continuity, and construction collaboration from a global perspective has become a pressing issue to be addressed in the in-depth application of BIM technology. Summary of the Invention

[0004] To address the problem that existing BIM model project partitioning relies on manual experience, single spatial or professional rules, making it difficult to coordinate component spatial proximity, system topology, and construction sequence relationships, and easily leading to logical breaks in critical system paths, this invention provides, in a first aspect, a BIM-based project-assisted partitioning method, comprising the following steps: Obtain geometric information, system information, and construction information of each component in the BIM model, construct a multi-dimensional feature vector containing normalized spatial coordinates, system topology coding, and construction process sequence, calculate the spatial distance, topological correlation, and construction proximity between components based on the multi-dimensional feature vector, and calculate the comprehensive connection strength by weighted fusion using adjustable weight parameters to generate a global connection strength matrix. Centered on the spatial coordinates of each component, an anisotropic potential field basis function with morphology adjusted by the global connection strength matrix is ​​constructed. All potential field basis functions are linearly superimposed to form a scalar potential field covering the model space. The scalar potential field is inverted to form an inverted scalar potential field. The local minimum point set with zero gradient and positive Hessian matrix is ​​solved as the cluster center. The watershed surface between adjacent cluster centers is extracted using the three-dimensional watershed algorithm as the initial segmentation boundary to obtain the initial partitioning scheme. The initial partitioning scheme is verified based on the critical system path integrity constraints based on preset truncation nodes. If logical breakpoints exist, the feature parameters of the components on both sides of the logical breakpoint, as well as the cluster labels of the current partitioned regions of the components on both sides and the spatial coordinates of the corresponding cluster centers, are extracted as breakpoint samples. The adjustable weight parameters are adjusted using a heuristic optimization algorithm, and the steps of calculating the spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vectors are iterated until the critical system path integrity constraints are met, and the working set partitioning result is output.

[0005] Optionally, the construction of a multi-dimensional feature vector comprising normalized spatial coordinates, system topology coding, and construction process sequence includes: Extract the center position of the bounding box of each component, subtract the minimum extreme value of the corresponding dimension of the model from the coordinate values ​​of each dimension of the center position, and divide by the range value of the dimension to obtain the normalized spatial coordinates. Read the hierarchical path of each component in the project classification tree, fill and splice it with hexadecimal code of uniform length in top-down order to generate the system topology code of the same length; Obtain the planned start time of each component in the construction schedule, calculate the absolute value of the time difference between the planned start time and the total start time of the project as the number of days, and generate the construction process sequence. The normalized spatial coordinates, the numerically mapped system topology code, and the construction process sequence are sequentially concatenated into a numerical one-dimensional array to form the multidimensional feature vector.

[0006] Optionally, calculating the spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vector includes: The Euclidean distance between any two components is calculated using their normalized spatial coordinates and is taken as the spatial distance. The system topology codes of any two components are compared from left to right, and the number of consecutive identical characters starting from the first character is counted. The ratio of the number of characters to the total length of the code is used as the topology correlation degree. Calculate the absolute value of the difference between the construction sequence of any two components, divide the absolute value by a preset time constant, and take the natural exponent of the opposite of the quotient as the construction proximity.

[0007] Optionally, the step of calculating the comprehensive connection strength through weighted fusion using adjustable weight parameters to generate a global connection strength matrix includes: Initialize three independent adjustable weight parameters: spatial weight, topological weight, and construction weight, and constrain the sum of the three to be one; The spatial distance between any two components is added to a preset bias constant, the reciprocal is taken, and normalization is performed to obtain the spatial proximity. The spatial proximity, the topological correlation, and the construction proximity are multiplied by the corresponding adjustable weight parameters and summed to obtain the comprehensive connection strength between the two components. Traverse all component combinations in the BIM model, calculate the comprehensive connection strength between each pair of components, map the component index to the corresponding position in the matrix, and generate the global connection strength matrix.

[0008] Optionally, the construction of anisotropic potential field basis functions, centered on the spatial coordinates of each component and whose shape is adjusted by the global connectivity strength matrix, includes: For any component, based on the row vector corresponding to the component in the global connection strength matrix, after excluding the matrix elements corresponding to the component itself, the target component with the largest comprehensive connection strength with the component is found; The spatial vector pointing from the component to the target component is taken as the principal axis direction of the potential field basis function; A three-dimensional Gaussian function is used as the basis function of the potential field. The variance of the function in the principal axis direction is set to a positive value that is positively correlated with the maximum comprehensive connection strength of the component, and the variance in the orthogonal direction of the Gaussian function is set to a preset constant.

[0009] Optionally, the step of verifying the initial partitioning scheme based on the critical system path integrity constraints of the preset truncation node, and if a logical breakpoint exists, extracting the feature parameters of the components on both sides of the logical breakpoint, as well as the clustering labels of the current segmented regions to which the components on both sides belong and the spatial coordinates of the corresponding cluster centers as breakpoint samples, includes: Extract the set of components in the BIM model that belong to the same electromechanical pipeline system, are physically connected, and do not contain preset cut-off nodes to form the critical system path; Check whether all components in the same critical system path are assigned to the same cluster label; If there are adjacent components belonging to the same critical system path but assigned to different cluster labels, then the connection point of the adjacent components is determined to be a logical breakpoint. The feature parameters of the components on both sides of the logical break point, as well as the cluster labels of the segmented regions to which the components on both sides currently belong and the spatial coordinates of the corresponding cluster centers, are extracted together as break samples.

[0010] Secondly, the present invention also provides a project-aided partitioning system based on BIM technology, comprising the following modules: The generation module is used to acquire the geometric information, system information and construction information of each component of the BIM model, construct a multi-dimensional feature vector containing normalized spatial coordinates, system topology coding and construction process sequence, calculate the spatial distance, topological correlation degree and construction proximity between components based on the multi-dimensional feature vector, and calculate the comprehensive connection strength by weighted fusion through adjustable weight parameters to generate a global connection strength matrix. The construction module is used to construct anisotropic potential field basis functions whose shape is adjusted by the global connection strength matrix, with the spatial coordinates of each component as the center. All potential field basis functions are linearly superimposed to form a scalar potential field covering the model space. The scalar potential field is inverted to form an inverted scalar potential field. The local minimum point set with zero gradient and positive Hessian matrix is ​​solved as the cluster center. The watershed surface between adjacent cluster centers is extracted using the three-dimensional watershed algorithm as the initial segmentation boundary to obtain the initial partitioning scheme. The partitioning module is used to verify the initial partitioning scheme based on the critical system path integrity constraints based on preset truncation nodes. If there are logical breakpoints, the feature parameters of the components on both sides of the logical breakpoint, as well as the cluster labels of the current partitioned regions of the components on both sides and the spatial coordinates of the corresponding cluster centers are extracted as breakpoint samples. The adjustable weight parameters are adjusted using a heuristic optimization algorithm, and the steps of calculating the spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vector are iterated until the critical system path integrity constraints are met, and the working set partitioning result is output.

[0011] Furthermore, the construction of a multi-dimensional feature vector comprising normalized spatial coordinates, system topology coding, and construction process sequence includes: Extract the center position of the bounding box of each component, subtract the minimum extreme value of the corresponding dimension of the model from the coordinate values ​​of each dimension of the center position, and divide by the range value of the dimension to obtain the normalized spatial coordinates. Read the hierarchical path of each component in the project classification tree, fill and splice it with hexadecimal code of uniform length in top-down order to generate the system topology code of the same length; Obtain the planned start time of each component in the construction schedule, calculate the absolute value of the time difference between the planned start time and the total start time of the project as the number of days, and generate the construction process sequence. The normalized spatial coordinates, the numerically mapped system topology code, and the construction process sequence are sequentially concatenated into a numerical one-dimensional array to form the multidimensional feature vector.

[0012] Furthermore, the calculation of spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vector includes: The Euclidean distance between any two components is calculated using their normalized spatial coordinates and is taken as the spatial distance. The system topology codes of any two components are compared from left to right, and the number of consecutive identical characters starting from the first character is counted. The ratio of the number of characters to the total length of the code is used as the topology correlation degree. Calculate the absolute value of the difference between the construction sequence of any two components, divide the absolute value by a preset time constant, and take the natural exponent of the opposite of the quotient as the construction proximity.

[0013] Furthermore, the step of calculating the comprehensive connection strength through weighted fusion using adjustable weight parameters to generate a global connection strength matrix includes: Initialize three independent adjustable weight parameters: spatial weight, topological weight, and construction weight, and constrain the sum of the three to be one; The spatial distance between any two components is added to a preset bias constant, the reciprocal is taken, and normalization is performed to obtain the spatial proximity. The spatial proximity, the topological correlation, and the construction proximity are multiplied by the corresponding adjustable weight parameters and summed to obtain the comprehensive connection strength between the two components. Traverse all component combinations in the BIM model, calculate the comprehensive connection strength between each pair of components, map the component index to the corresponding position in the matrix, and generate the global connection strength matrix.

[0014] Furthermore, the construction of anisotropic potential field basis functions, whose shape is adjusted by the global connectivity strength matrix, centered on the spatial coordinates of each component, includes: For any component, based on the row vector corresponding to the component in the global connection strength matrix, after excluding the matrix elements corresponding to the component itself, the target component with the largest comprehensive connection strength with the component is found; The spatial vector pointing from the component to the target component is taken as the principal axis direction of the potential field basis function; A three-dimensional Gaussian function is used as the basis function of the potential field. The variance of the function in the principal axis direction is set to a positive value that is positively correlated with the maximum comprehensive connection strength of the component, and the variance in the orthogonal direction of the Gaussian function is set to a preset constant.

[0015] Further, the initial partitioning scheme is validated based on the critical system path integrity constraints of the preset truncation node. If a logical breakpoint exists, the feature parameters of the components on both sides of the logical breakpoint, as well as the cluster labels of the current partitioned regions to which the components on both sides belong and the spatial coordinates of the corresponding cluster centers, are extracted as breakpoint samples, including: Extract the set of components in the BIM model that belong to the same electromechanical pipeline system, are physically connected, and do not contain preset cut-off nodes to form the critical system path; Check whether all components in the same critical system path are assigned to the same cluster label; If there are adjacent components belonging to the same critical system path but assigned to different cluster labels, then the connection point of the adjacent components is determined to be a logical breakpoint. The feature parameters of the components on both sides of the logical break point, as well as the cluster labels of the segmented regions to which the components on both sides currently belong and the spatial coordinates of the corresponding cluster centers, are extracted together as break samples.

[0016] This invention extracts multi-dimensional features of BIM model components, including spatial dimensions, system topology, and construction sequence, and weights and fuses them to generate a global connectivity strength matrix to characterize the comprehensive relationships between components. Based on this, an anisotropic potential field is constructed and a three-dimensional watershed algorithm is used to extract the partitioning boundary, enabling the correlation of discrete components to be transformed into a continuous spatial partitioning model, which improves the rationality of the working set partitioning. Furthermore, closed-loop verification is performed through critical system path integrity constraints. When logical breaks are detected, a heuristic optimization algorithm is used to adjust the weight parameters, reducing the probability of pipelines or critical paths being unreasonably truncated, thereby improving the accuracy and engineering applicability of project auxiliary partitioning. Attached Figure Description

[0017] Figure 1 A flowchart of a project partitioning method based on BIM technology; Figure 2 This is a schematic diagram illustrating the relationship between construction proximity and time difference. Detailed Implementation

[0018] 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 some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] This specification provides one or more embodiments of a project-aided partitioning method based on BIM technology, such as... Figure 1 As shown, it includes the following steps: S1. Obtain the geometric information, system information, and construction information of each component in the BIM model, construct a multi-dimensional feature vector containing normalized spatial coordinates, system topology coding, and construction process sequence, and calculate the spatial distance, topological correlation, and construction proximity between components based on the multi-dimensional feature vector. Calculate the comprehensive connection strength by weighting and fusing the components using adjustable weight parameters, and generate a global connection strength matrix.

[0020] The target components in the BIM model are obtained, and the center 3D coordinates of the bounding box of each component are extracted as geometric information. The physical connections between MEP (Mechanical, Electrical, and Plumbing) pipelines and equipment are also obtained as system information. The hierarchical paths of each component in the project classification tree are read, and a uniform-length hexadecimal system topology code is generated in a top-down order. The planned start date of each component in the schedule is read, and the time difference in days between this date and the total project start date is calculated as the construction sequence in days. The 3D coordinates of the component center are normalized using minimum and maximum values ​​to obtain normalized spatial coordinates. The normalized spatial coordinates, the numerically mapped system topology code, and the construction sequence are concatenated column-wise to form a single-column real number array, serving as the multi-dimensional feature vector of the component. Simultaneously, the original fixed-length hexadecimal system topology code is retained separately as a topology comparison field.

[0021] For any two components in the model, the Euclidean distance between the normalized spatial coordinates of the two multidimensional feature vectors is calculated as the spatial distance. The original fixed-length hexadecimal system topology codes of any two components are compared digit by digit, starting from the first character. The number of consecutive identical characters is counted, and this number is divided by the total code length to obtain the topology correlation degree. The absolute value of the difference in the construction process order of the two components is calculated and converted using a natural exponential decay function to obtain the construction proximity degree. The spatial distance weight, topology correlation degree weight, and construction proximity degree weight are initialized in memory as adjustable weight parameters. Spatial proximity is obtained by adding a preset bias constant to the spatial distance, taking its reciprocal, and normalizing it. Then, the spatial proximity is multiplied by the spatial distance weight, the topological correlation weight, and the construction proximity weight. These three multiplications are summed to obtain the comprehensive connection strength between each pair of components. A two-dimensional floating-point array of all zeros is allocated in memory based on the total number of components. A mapping table between unique component identifiers and continuous matrix indices is established. The comprehensive connection strength value between each pair of components is then sequentially filled into the corresponding row and column positions using the continuous matrix indices to generate a two-dimensional global connection strength matrix. The relationship between construction proximity and time difference is shown in the curve. Figure 2 As shown.

[0022] In some embodiments, the construction of a multidimensional feature vector comprising normalized spatial coordinates, system topology coding, and construction process sequence includes: Extract the center position of the bounding box of each component, subtract the minimum extreme value of the corresponding dimension of the model from the coordinate values ​​of each dimension of the center position, and divide by the range value of the dimension to obtain the normalized spatial coordinates. Read the hierarchical path of each component in the project classification tree, fill and splice it with hexadecimal code of uniform length in top-down order to generate the system topology code of the same length; Obtain the planned start time of each component in the construction schedule, calculate the absolute value of the time difference between the planned start time and the total start time of the project as the number of days, and generate the construction process sequence. The normalized spatial coordinates, the numerically mapped system topology code, and the construction process sequence are sequentially concatenated into a numerical one-dimensional array to form the multidimensional feature vector.

[0023] By using secondary development interfaces of the BIM model, such as Revit API, all three-dimensional solid components are traversed, the axial bounding box (AABB) of each component is read, and the average of the maximum and minimum values ​​of the X, Y, and Z axis coordinates is calculated. This average three-axis coordinate is used as the center position coordinate. The global minimum coordinate point (-50000, -50000, -50000) mm and the global range (100000 mm) of the entire BIM model space are obtained. The coordinate values ​​are linearly mapped to a continuous floating-point range of 0 to 1 using the minimum-maximum normalization formula to reduce the numerical calculation deviation caused by the large-scale spatial span. In terms of topology and construction information processing, based on the classification standard tree, such as the four-level structure of profession-system-subsystem-equipment, each level is assigned a code of preset length. For example, "HVAC-Air Conditioning Water-Water Supply-Valves" is sequentially converted into 2-digit hexadecimal codes 01, 0A, 03, and 1F, and concatenated from top to bottom into a fixed 8-digit string "010A031F". For components with insufficient levels, "00" is added to the end. The baseline start time in the project-level BIM schedule management file is set to... Read the planned start time of a certain component The number of days between them, for example, 60 days, is calculated as a time sequence feature.

[0024] To ensure computability across different data types, the hexadecimal system topology code is converted into its corresponding decimal value, such as "010A031F" becoming 17433375. This decimal value is then sequentially combined with normalized spatial coordinates in three dimensions (e.g., (0.6, 0.45, 0.8)) and the number of construction days in one dimension (e.g., 60) to generate a one-dimensional floating-point array of length 5 for each discrete component in the model. The fixed-length hexadecimal system topology code is retained separately as a topology comparison field for subsequent character-by-character comparison. This multidimensional feature vector serves as the data foundation, enabling mathematical operations on features related to spatial location, business logic, and time dimensions.

[0025] In some embodiments, calculating the spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vector includes: The Euclidean distance between any two components is calculated using their normalized spatial coordinates and is taken as the spatial distance. The system topology codes of any two components are compared from left to right, and the number of consecutive identical characters starting from the first character is counted. The ratio of the number of characters to the total length of the code is used as the topology correlation degree. Calculate the absolute value of the difference between the construction sequence of any two components, divide the absolute value by a preset time constant, and take the natural exponent of the opposite of the quotient as the construction proximity.

[0026] When representing multidimensional features, the normalized spatial coordinates between 0 and 1 in the first three bits of the multidimensional feature vectors of any two components are extracted. The square root of the sum of squares is then calculated using the standard three-dimensional Euclidean distance formula. The resulting spatial distance distribution is within a continuous interval of 0 to 1.732 (√3), representing the degree of spatial dispersion. For the original 8-bit hexadecimal system topology encoding before decimal conversion, such as “010A031F” for component A and “010A032B” for component B, a string matching mechanism is used. Starting from the first index, each character is traversed and compared to the next, until the first mismatched character is encountered. In the example above, the consecutive identical characters include “0,1,0,A,0,3” (6 characters in total). Dividing this number by the preset total length of 8 yields a topology correlation of 0.75. This parameter, between 0 and 1, indicates the association between the two components in the electromechanical system or building structure classification tree.

[0027] In the time-dimensional calculation, the planned start days of the last two component vectors are read, and the absolute difference in days after subtracting the two is calculated. To convert the divergent construction period days into a normalized weighting coefficient between 0 and 1, a preset time decay constant is introduced, typically preferred to be between 15 and 45 days. An example value is set to 30 days, representing a standard monthly construction sequence. The absolute difference is divided by 30 days, and the negative of the quotient is taken as the exponent of the natural logarithm e. For example, when the difference in days is 15 days, the construction proximity is approximately 0.606. This exponential decay model can simulate the clustering effect of construction progress, so that components with closer start times obtain higher proximity, while the proximity of components with a time span exceeding one sequence will decay, ensuring that the calculated three core parameters are within a comparable numerical range and conform to engineering logic.

[0028] In some embodiments, the step of calculating the comprehensive connectivity strength by weighted fusion using adjustable weight parameters to generate a global connectivity strength matrix includes: Initialize three independent adjustable weight parameters: spatial weight, topological weight, and construction weight, and constrain the sum of the three to be one; The spatial distance between any two components is added to a preset bias constant, the reciprocal is taken, and normalization is performed to obtain the spatial proximity. The spatial proximity, the topological correlation, and the construction proximity are multiplied by the corresponding adjustable weight parameters and summed to obtain the comprehensive connection strength between the two components. Traverse all component combinations in the BIM model, calculate the comprehensive connection strength between each pair of components, map the component index to the corresponding position in the matrix, and generate the global connection strength matrix.

[0029] A set of hyperparameters floating between 0 and 1 is pre-configured. For example, the spatial weight, topological weight, and construction weight are set to 0.3, 0.4, and 0.3 respectively, with a normalized boundary condition restricting their sum to 1.0. For the extracted spatial distance d, a very small bias constant, preferably 0.01, is added when calculating the reciprocal of the spatial distance. The resulting reciprocal term is then normalized to obtain the spatial proximity. This spatial proximity is multiplied by the spatial weight, and then the product of topological correlation and topological weight, and the product of construction proximity and construction weight are added respectively to obtain the comprehensive connection strength value. This process reduces the risk of excessive amplification of the spatial term due to extremely close components, allowing all three dimensions—space, topology, and construction—to participate in the connection strength calculation.

[0030] A double-nested loop is constructed to traverse the entire BIM model set containing N components. The weighted fusion formula described above is used to calculate the comprehensive connection strength value for all N×(N-1) / 2 pairs of non-repeating component combinations. A symmetric dense floating-point two-dimensional matrix of size N×N is instantiated. A continuous index mapping table is established using the unique identifier of each component, mapping each component to a continuous matrix index in the range of 0 to N-1. The calculated comprehensive connection strength is then filled into the corresponding (i,j) coordinate points. For the diagonal element (i,i) of the matrix, since it is related to the component itself, a globally maximum preset peak constant is assigned to this element. This global connection strength matrix constitutes the numerical basis for the subsequent generation of the principal axis orientation and shape of the anisotropic potential field basis functions, and excludes the diagonal elements corresponding to the component itself when searching for target components.

[0031] S2. Using the spatial coordinates of each component as the center, construct an anisotropic potential field basis function whose shape is adjusted by the global connection strength matrix. Linearly superimpose all potential field basis functions to form a scalar potential field covering the model space. Invert the scalar potential field to form an inverted scalar potential field. Solve for the local minimum point set with zero gradient and positive Hessian matrix as the cluster center. Use the three-dimensional watershed algorithm to extract the watershed surface between adjacent cluster centers as the initial segmentation boundary to obtain the initial partitioning scheme.

[0032] A three-dimensional anisotropic Gaussian kernel function is established as the potential field basis function, centered on the normalized spatial coordinates of each component. The comprehensive connection strength between the component and other surrounding components (excluding itself) is extracted based on the global connection strength matrix. For any component, after excluding its own matrix elements, the target component with the largest comprehensive connection strength is found. The spatial vector pointing from the current component to the target component is taken as the principal axis direction of the potential field basis function. A three-dimensional Gaussian function is used as the potential field basis function, and the variance of the Gaussian function along the principal axis direction is set to a positive value directly correlated with the maximum comprehensive connection strength, while the variance along the orthogonal direction is set to a preset constant. This allows the potential field to have a larger coverage scale along the strong connection direction, thereby forming an anisotropic local potential field distribution.

[0033] A set of three-dimensional regular mesh points with a specified spatial resolution is generated within the three-dimensional bounding box space of the model. All components are traversed, and continuous function values ​​of the three-dimensional anisotropic Gaussian kernel function corresponding to all components are calculated at each three-dimensional mesh point. All Gaussian kernel function values ​​at the same three-dimensional mesh point are summed to generate a scalar potential field covering the three-dimensional mesh space of the model. The value of each element in the scalar potential field is multiplied by -1 to obtain the inverted scalar potential field.

[0034] The first-order partial derivatives of the three-dimensional inverted scalar potential field are calculated along the three spatial coordinate axes to obtain the three-dimensional gradient field data. The first-order partial derivatives of the three-dimensional gradient field data are then calculated to obtain the three-dimensional Hessian matrix field data. Each three-dimensional grid point is traversed, and the L2 norm of the three-dimensional gradient value vector is set to less than 10. -6 The grid points are identified as candidate points, and the three eigenvalues ​​of the Hessian matrix corresponding to each candidate point are calculated. When all three eigenvalues ​​are greater than zero, the corresponding candidate point is identified as a local minimum point with a gradient approximately zero and a positive definite Hessian matrix, and this local minimum point is used as the cluster center. The three-dimensional inverted scalar potential field is used as the segmentation input, and the cluster centers are used as segmentation seed points. Three-dimensional watershed segmentation is performed in the three-dimensional grid space to obtain a three-dimensional region partitioning array that assigns segmentation labels to each grid point. The interface between adjacent grid points with different labels is extracted from the three-dimensional region partitioning array, and a spatial watershed surface is constructed as the initial segmentation boundary. Based on the grid space sub-region where the three-dimensional geometric center of each component is located, all components are assigned to the corresponding cluster center label cluster to form the initial partitioning scheme.

[0035] In some embodiments, the construction of anisotropic potential field basis functions, whose shape is adjusted by the global connectivity strength matrix, centered on the spatial coordinates of each component, includes: For any component, based on the row vector corresponding to the component in the global connection strength matrix, after excluding the matrix elements corresponding to the component itself, the target component with the largest comprehensive connection strength with the component is found; The spatial vector pointing from the component to the target component is taken as the principal axis direction of the potential field basis function; A three-dimensional Gaussian function is used as the basis function of the potential field. The variance of the function in the principal axis direction is set to a positive value that is positively correlated with the maximum comprehensive connection strength of the component, and the variance in the orthogonal direction of the Gaussian function is set to a preset constant.

[0036] In the process of generating the scalar potential field, any component in the three-dimensional space of the model is denoted as the origin component. Extract the i-th row of the one-dimensional array from the aforementioned global connectivity strength matrix, remove the diagonal elements, and then iterate through the array to find the extreme values, locating the elements that are connected to the global connectivity strength matrix. Extract the target component corresponding to column index j with the highest connection strength value. Normalized three-dimensional spatial coordinates. The coordinates are the starting point, The coordinates of the vectors are used as the endpoints for vector subtraction, and the result is normalized to obtain a three-dimensional unit space vector with a definite direction. This vector is then set as the covering vector. The principal axis direction of the basis function of the anisotropic potential field in the local area of ​​the point is used to cause the originally isotropic potential field to extend directionally along the direction of the high-connection strength component.

[0037] In the mathematical modeling of the potential field morphology, the standard three-dimensional Gaussian function distribution formula is selected as the kernel of the potential field basis functions. To achieve the mapping between the principal axis extension and the comprehensive connection strength, a positively correlated conversion formula is constructed, wherein the connection strength adjustment term is calculated based on the normalized value of the maximum comprehensive connection strength. Specifically, the maximum comprehensive connection strength corresponding to the current component is denoted as... Normalization is performed to obtain the normalized connection strength. Then adjust the preset variance range and... Multiplying these yields the connection strength adjustment term. The variance along the principal axis is equal to the sum of the base variance and the connection strength adjustment term. Therefore, the greater the maximum overall connection strength, the larger the connection strength adjustment term, the greater the variance along the principal axis, and the correspondingly larger the coverage area of ​​the potential field along the strong connection direction. Simultaneously, the variances along the local X-axis and Y-axis (two orthogonal plane directions) are set to preset constants. This variance configuration generates a directional potential field around each component, allowing strongly connected component pairs to form a more continuous potential field distribution through linear superposition, which drives watershed clustering.

[0038] S3. The initial partitioning scheme is verified based on the critical system path integrity constraint based on the preset truncation node. If there is a logical breakpoint, the feature parameters of the components on both sides of the logical breakpoint, as well as the cluster labels of the current partitioned regions of the components on both sides and the spatial coordinates of the corresponding cluster centers are extracted as breakpoint samples. The adjustable weight parameters are adjusted using a heuristic optimization algorithm and the steps of calculating the spatial distance, topological correlation and construction proximity between components based on the multidimensional feature vector are iterated until the critical system path integrity constraint is met, and the working set partitioning result is output.

[0039] Based on a depth-first traversal method, components belonging to the main ventilation duct and pressurized water supply and drainage network system are obtained, forming a critical system path topology sequence. Adjacent connecting components in this topology sequence that have been assigned to different sub-region labels by the initial partitioning scheme are then detected. It is determined whether the connection between the adjacent connecting components contains preset cut-off nodes such as flanges or flexible joints. If the connection contains a preset cut-off node, the components on both sides are allowed to be partitioned into different sub-regions. If the connection does not contain a preset cut-off node and the components on both sides belong to different sub-regions, the connection is determined not to meet the critical system path integrity constraint, and the location of the corresponding adjacent connecting component is marked as a logical breakpoint.

[0040] If at least one logical breakpoint is found after global detection, the multi-dimensional feature vectors of the electromechanical pipeline components on both sides of the logical breakpoint are extracted. Based on the three-dimensional region partitioning array generated by the three-dimensional watershed segmentation, the labels of the currently assigned segmented regions of the components on both sides are determined and used as the cluster labels for the components. Simultaneously, the spatial coordinates of the corresponding cluster centers are extracted and merged to construct a breakpoint sample dataset representing the segmentation anomaly. Based on the total number of logical breakpoints in the current initial partitioning scheme and the distance penalty term from the component to its current cluster center in the breakpoint sample dataset, a weighted sum is performed to obtain the breakpoint penalty fitness value, which is used as the objective function of the heuristic optimization algorithm. The cluster labels are used to count the number of logical breakpoints in the critical system paths, and the spatial coordinates of the cluster centers are used to calculate the distance penalty term between the components on both sides of the breakpoint and their current cluster centers.

[0041] Spatial distance weight, topological correlation weight, and construction proximity weight are used as three-dimensional optimization decision variables, and new adjustable weight parameter combinations are generated within a preset value range of 0 to 1. Each newly generated weight parameter is normalized to ensure that the sum of the spatial distance weight, topological correlation weight, and construction proximity weight remains one. The new weight parameter combination overwrites the old weight parameters, and the process returns to the step of calculating the comprehensive connection strength between components, regenerating the global connection strength matrix, the three-dimensional inverted scalar potential field, and the initial partitioning scheme.

[0042] The critical system path integrity verification and weight iterative optimization process is repeated until any of the following stopping conditions are met: First, the number of logical breakpoints in the current partitioning scheme is zero; second, the relative decrease in the fitness value of the optimal breakpoint penalty in several consecutive iterations is less than a preset convergence threshold; third, the number of iterations reaches a preset iteration limit. Preferably, the preset iteration limit is fifty iterations, and the preset convergence condition is that the relative decrease in the fitness value of the optimal breakpoint penalty in five consecutive iterations is less than one percent. If logical breakpoints that do not meet the constraints still exist after stopping the iteration, local merging is performed on continuous components of the same critical system path that do not contain preset truncation nodes, so that they are included in the same working set. Finally, all components in the model are traversed, the clustering identifier of the segmented sub-region of each component is extracted as the working set partitioning identifier attribute value, and the working set partitioning identifier attribute value is written into the system shared parameter table of the BIM model, and the working set partitioning result is output.

[0043] In some embodiments, the step of verifying the initial partitioning scheme based on the critical system path integrity constraints of preset truncation nodes, and if logical breakpoints exist, extracting the feature parameters of the components on both sides of the logical breakpoint, as well as the clustering labels of the current segmented regions to which the components on both sides belong and the spatial coordinates of the corresponding cluster centers as breakpoint samples, includes: Extract the set of components in the BIM model that belong to the same electromechanical pipeline system, are physically connected, and do not contain preset cut-off nodes to form the critical system path; Check whether all components in the same critical system path are assigned to the same cluster label; If there are adjacent components belonging to the same critical system path but assigned to different cluster labels, then the connection point of the adjacent components is determined to be a logical breakpoint. The feature parameters of the components on both sides of the logical break point, as well as the cluster labels of the segmented regions to which the components on both sides currently belong and the spatial coordinates of the corresponding cluster centers, are extracted together as break samples.

[0044] After obtaining the initial segmentation boundary based on the watershed algorithm, the business logic layer verification is initiated. It traverses the MEP (Mechanical, Electrical, and Plumbing) systems in the BIM model, such as HVAC or fire water systems, and sets various valves, equipment interfaces, flexible joints, flanges, or tees where main and branch pipes intersect as preset cutoff nodes. A depth-first search is performed along the connecting ports of the pipelines, encapsulating ordinary straight pipe sections and elbows connected end-to-end between two cutoff nodes into critical system paths. The initial clustering ID labels of each component in these critical system path sets are compared, and a consistency test is performed. If an ID inconsistency is found within the same path, for example, pipe segment A has an ID in cluster 1, while the closely connected pipe segment B has an ID in cluster 2, the verification fails, and the connection interface between pipe segment A and pipe segment B that does not contain a preset cutoff node is marked at the geometric level as a logical breakpoint that disrupts system integrity.

[0045] After completing consistency testing of all critical system paths, the total number of logical breakpoints in the current initial partitioning scheme is counted. If the total number of logical breakpoints is greater than zero, a penalty correction process is initiated. For each logical breakpoint, components A and B on both sides of the breakpoint are extracted. Based on the region partitioning results generated by the 3D watershed segmentation, the cluster labels corresponding to the current segmented regions of components A and B, as well as the spatial coordinates of the corresponding cluster centers in the model space, are found. Component A, component B, and their current cluster center spatial coordinates and cluster labels are grouped as a breakpoint sample set. All breakpoint sample sets are then aggregated to form a breakpoint sample dataset.

[0046] When calculating the iterative fitness function value, a fracture quantity penalty term is constructed based on the total number of logical fracture points, and a distance penalty term is constructed based on the distance between the components on both sides of each logical fracture point and their current cluster centers. The fracture quantity penalty term and the distance penalty term are then weighted and fused to obtain the fracture penalty fitness value. The fracture penalty fitness value is input into the heuristic optimization algorithm, enabling the optimizer to adjust the spatial weights, topological weights, and construction weights in the next iteration, and to update the global connectivity strength matrix and potential field distribution through closed-loop weight feedback.

[0047] This specification also provides one or more embodiments of a project-aided partitioning system based on BIM technology, including the following modules: The generation module is used to acquire the geometric information, system information and construction information of each component of the BIM model, construct a multi-dimensional feature vector containing normalized spatial coordinates, system topology coding and construction process sequence, calculate the spatial distance, topological correlation degree and construction proximity between components based on the multi-dimensional feature vector, and calculate the comprehensive connection strength by weighted fusion through adjustable weight parameters to generate a global connection strength matrix. The construction module is used to construct anisotropic potential field basis functions whose shape is adjusted by the global connection strength matrix, with the spatial coordinates of each component as the center. All potential field basis functions are linearly superimposed to form a scalar potential field covering the model space. The scalar potential field is inverted to form an inverted scalar potential field. The local minimum point set with zero gradient and positive Hessian matrix is ​​solved as the cluster center. The watershed surface between adjacent cluster centers is extracted using the three-dimensional watershed algorithm as the initial segmentation boundary to obtain the initial partitioning scheme. The partitioning module is used to verify the initial partitioning scheme based on the critical system path integrity constraints based on preset truncation nodes. If there are logical breakpoints, the feature parameters of the components on both sides of the logical breakpoint, as well as the cluster labels of the current partitioned regions of the components on both sides and the spatial coordinates of the corresponding cluster centers are extracted as breakpoint samples. The adjustable weight parameters are adjusted using a heuristic optimization algorithm, and the steps of calculating the spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vector are iterated until the critical system path integrity constraints are met, and the working set partitioning result is output.

[0048] The above description represents the preferred embodiments of the present invention. It should be noted that, for those skilled in the art, various improvements and modifications can be made without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A project-aided partitioning method based on BIM technology, characterized in that, Includes the following steps: Obtain geometric information, system information, and construction information of each component in the BIM model, construct a multi-dimensional feature vector containing normalized spatial coordinates, system topology coding, and construction process sequence, calculate the spatial distance, topological correlation, and construction proximity between components based on the multi-dimensional feature vector, and calculate the comprehensive connection strength by weighted fusion using adjustable weight parameters to generate a global connection strength matrix. Centered on the spatial coordinates of each component, an anisotropic potential field basis function with morphology adjusted by the global connection strength matrix is ​​constructed. All potential field basis functions are linearly superimposed to form a scalar potential field covering the model space. The scalar potential field is inverted to form an inverted scalar potential field. The local minimum point set with zero gradient and positive Hessian matrix is ​​solved as the cluster center. The watershed surface between adjacent cluster centers is extracted using the three-dimensional watershed algorithm as the initial segmentation boundary to obtain the initial partitioning scheme. The initial partitioning scheme is verified based on the critical system path integrity constraints based on preset truncation nodes. If logical breakpoints exist, the feature parameters of the components on both sides of the logical breakpoint, as well as the cluster labels of the current partitioned regions of the components on both sides and the spatial coordinates of the corresponding cluster centers, are extracted as breakpoint samples. The adjustable weight parameters are adjusted using a heuristic optimization algorithm, and the steps of calculating the spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vectors are iterated until the critical system path integrity constraints are met, and the working set partitioning result is output.

2. The method according to claim 1, characterized in that, The construction of a multi-dimensional feature vector, comprising normalized spatial coordinates, system topology coding, and construction process sequence, includes: Extract the center position of the bounding box of each component, subtract the minimum extreme value of the corresponding dimension of the model from the coordinate values ​​of each dimension of the center position, and divide by the range value of the dimension to obtain the normalized spatial coordinates. Read the hierarchical path of each component in the project classification tree, fill and splice it with hexadecimal code of uniform length in top-down order to generate the system topology code of the same length; Obtain the planned start time of each component in the construction schedule, calculate the absolute value of the time difference between the planned start time and the total start time of the project as the number of days, and generate the construction process sequence. The normalized spatial coordinates, the numerically mapped system topology code, and the construction process sequence are sequentially concatenated into a numerical one-dimensional array to form the multidimensional feature vector.

3. The method according to claim 2, characterized in that, The calculation of spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vector includes: The Euclidean distance between any two components is calculated using their normalized spatial coordinates and is taken as the spatial distance. The system topology codes of any two components are compared from left to right, and the number of consecutive identical characters starting from the first character is counted. The ratio of the number of characters to the total length of the code is used as the topology correlation degree. Calculate the absolute value of the difference between the construction sequence of any two components, divide the absolute value by a preset time constant, and take the natural exponent of the opposite of the quotient as the construction proximity.

4. The method according to claim 1, characterized in that, The step of calculating the comprehensive connection strength through weighted fusion using adjustable weight parameters to generate a global connection strength matrix includes: Initialize three independent adjustable weight parameters: spatial weight, topological weight, and construction weight, and constrain the sum of the three to be one; The spatial distance between any two components is added to a preset bias constant, the reciprocal is taken, and normalization is performed to obtain the spatial proximity. The spatial proximity, the topological correlation, and the construction proximity are multiplied by the corresponding adjustable weight parameters and summed to obtain the comprehensive connection strength between the two components. Traverse all component combinations in the BIM model, calculate the comprehensive connection strength between each pair of components, map the component index to the corresponding position in the matrix, and generate the global connection strength matrix.

5. The method according to claim 1, characterized in that, The anisotropic potential field basis functions, centered on the spatial coordinates of each component and whose shape is adjusted by the global connectivity strength matrix, include: For any component, based on the row vector corresponding to the component in the global connection strength matrix, after excluding the matrix elements corresponding to the component itself, the target component with the largest comprehensive connection strength with the component is found; The spatial vector pointing from the component to the target component is taken as the principal axis direction of the potential field basis function; A three-dimensional Gaussian function is used as the basis function of the potential field. The variance of the function in the principal axis direction is set to a positive value that is positively correlated with the maximum comprehensive connection strength of the component, and the variance in the orthogonal direction of the Gaussian function is set to a preset constant.

6. The method according to claim 1, characterized in that, The initial partitioning scheme is validated based on the critical system path integrity constraints of preset truncation nodes. If logical breakpoints exist, feature parameters of the components on both sides of the logical breakpoint, as well as the cluster labels of the current segmented regions to which the components on both sides belong and the spatial coordinates of the corresponding cluster centers, are extracted as breakpoint samples, including: Extract the set of components in the BIM model that belong to the same electromechanical pipeline system, are physically connected, and do not contain preset cut-off nodes to form the critical system path; Check whether all components in the same critical system path are assigned to the same cluster label; If there are adjacent components belonging to the same critical system path but assigned to different cluster labels, then the connection point of the adjacent components is determined to be a logical breakpoint. The feature parameters of the components on both sides of the logical break point, as well as the cluster labels of the segmented regions to which the components on both sides currently belong and the spatial coordinates of the corresponding cluster centers, are extracted together as break samples.

7. A project-aided partitioning system based on BIM technology, characterized in that, Includes the following modules: The generation module is used to acquire the geometric information, system information and construction information of each component of the BIM model, construct a multi-dimensional feature vector containing normalized spatial coordinates, system topology coding and construction process sequence, calculate the spatial distance, topological correlation degree and construction proximity between components based on the multi-dimensional feature vector, and calculate the comprehensive connection strength by weighted fusion through adjustable weight parameters to generate a global connection strength matrix. The construction module is used to construct anisotropic potential field basis functions whose shape is adjusted by the global connection strength matrix, with the spatial coordinates of each component as the center. All potential field basis functions are linearly superimposed to form a scalar potential field covering the model space. The scalar potential field is inverted to form an inverted scalar potential field. The local minimum point set with zero gradient and positive Hessian matrix is ​​solved as the cluster center. The watershed surface between adjacent cluster centers is extracted using the three-dimensional watershed algorithm as the initial segmentation boundary to obtain the initial partitioning scheme. The partitioning module is used to verify the initial partitioning scheme based on the critical system path integrity constraints based on preset truncation nodes. If there are logical breakpoints, the feature parameters of the components on both sides of the logical breakpoint, as well as the cluster labels of the current partitioned regions of the components on both sides and the spatial coordinates of the corresponding cluster centers are extracted as breakpoint samples. The adjustable weight parameters are adjusted using a heuristic optimization algorithm, and the steps of calculating the spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vector are iterated until the critical system path integrity constraints are met, and the working set partitioning result is output.

8. The system according to claim 7, characterized in that, The construction of a multi-dimensional feature vector, comprising normalized spatial coordinates, system topology coding, and construction process sequence, includes: Extract the center position of the bounding box of each component, subtract the minimum extreme value of the corresponding dimension of the model from the coordinate values ​​of each dimension of the center position, and divide by the range value of the dimension to obtain the normalized spatial coordinates. Read the hierarchical path of each component in the project classification tree, fill and splice it with hexadecimal code of uniform length in top-down order to generate the system topology code of the same length; Obtain the planned start time of each component in the construction schedule, calculate the absolute value of the time difference between the planned start time and the total start time of the project as the number of days, and generate the construction process sequence. The normalized spatial coordinates, the numerically mapped system topology code, and the construction process sequence are sequentially concatenated into a numerical one-dimensional array to form the multidimensional feature vector.

9. The system according to claim 7, characterized in that, The calculation of spatial distance, topological correlation, and construction proximity between components based on the multidimensional feature vector includes: The Euclidean distance between any two components is calculated using their normalized spatial coordinates and is taken as the spatial distance. The system topology codes of any two components are compared from left to right, and the number of consecutive identical characters starting from the first character is counted. The ratio of the number of characters to the total length of the code is used as the topology correlation degree. Calculate the absolute value of the difference between the construction sequence of any two components, divide the absolute value by a preset time constant, and take the natural exponent of the opposite of the quotient as the construction proximity.

10. The system according to claim 7, characterized in that, The step of calculating the comprehensive connection strength through weighted fusion using adjustable weight parameters to generate a global connection strength matrix includes: Initialize three independent adjustable weight parameters: spatial weight, topological weight, and construction weight, and constrain the sum of the three to be one; The spatial distance between any two components is added to a preset bias constant, the reciprocal is taken, and normalization is performed to obtain the spatial proximity. The spatial proximity, the topological correlation, and the construction proximity are multiplied by the corresponding adjustable weight parameters and summed to obtain the comprehensive connection strength between the two components. Traverse all component combinations in the BIM model, calculate the comprehensive connection strength between each pair of components, map the component index to the corresponding position in the matrix, and generate the global connection strength matrix.

Images