Building engineering visual management method and system based on BIM

Abstract

The invention discloses a BIM-based construction engineering visual management method and system, and relates to construction engineering management, and the method comprises the steps: collecting a data set containing geometric information, attribute information and associated information; dividing the data set into a plurality of hierarchical subsets according to the hierarchical relationship of the building engineering BIM model; an improved DBSCAN clustering algorithm is adopted to perform clustering analysis on each hierarchy subset; classifying and archiving the data set according to the data cluster, and generating a semantic association data collaboration array A according to a classifying and archiving result; carrying out dimension reduction processing on the semantically associated data collaboration array A; constructing a spatial index of the dimension reduction data array by adopting a density-based spatial index method; and dividing the building engineering BIM model into a plurality of partitions by using a distributed computing framework, and performing distributed matching on the partitions and the spatial index to obtain a data matching point set. For the problem that the BIM-based building engineering high-dimensional data clustering precision is low in the prior art, the analysis precision is improved.

Description

Technical Field

[0001] This application relates to the field of construction project management, and particularly to a BIM-based visual management method and system for construction projects. Background Art

[0002] With the increasing complexity and informatization of construction projects, Building Information Modeling (BIM) technology has been widely used in the construction field. BIM models can integrate geometric information, attribute information, and association information of construction projects, providing a digital platform and tools for the design, construction, and management of construction projects. However, construction project BIM models usually contain a large amount of high-dimensional and heterogeneous data. How to effectively organize, analyze, and utilize these data is the key to realizing visual management of construction projects.

[0003] Currently, there have been some studies and applications for the analysis and management of construction project BIM data. Among them, data clustering is an important data analysis technology. By classifying similar data points into the same category, the internal structure and association of the data can be discovered. Traditional clustering algorithms, such as K-means and hierarchical clustering, have some limitations when dealing with high-dimensional data, such as being sensitive to the initial values of clustering results and being difficult to discover non-convex-shaped clusters. To overcome these limitations, density-based clustering algorithms, such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise), have been widely used in high-dimensional data clustering. The DBSCAN algorithm can discover clusters of any shape by identifying the local density of data points and is robust to noise data. However, existing BIM data clustering methods based on DBSCAN still have some deficiencies. First, construction project BIM models contain multiple levels of data, such as buildings, floors, components, etc. Existing methods usually regard all data points as the same level, ignoring the hierarchical relationship of the data, resulting in low clustering accuracy. Second, the data attributes in BIM models have different importance. Existing methods fail to fully consider the weight differences of attributes, affecting the quality of clustering. In addition, due to the high-dimensionality and complexity of BIM data, the interpretation and application of clustering results also face challenges.

[0004] For example, the related patent document CN117787670B discloses a BIM data management method and system for construction projects, belonging to the field of data management. The method includes: collecting data points from multiple data sources within the target construction project to generate a construction project data set; performing clustering analysis on the historical construction project data of the target construction project to generate multiple construction project data classes; generating a data collaboration array, performing dimensionality reduction processing to obtain a dimensionality-reduced data array, traversing the BIM data of the target construction project to match with the dimensionality-reduced data array, obtaining a data matching point set, performing analysis and management of the BIM data, and generating a management strategy; managing the BIM data of the target construction project through the management strategy. However, this solution only performs clustering analysis on the historical data of the construction project, fails to fully utilize the geometric information, attribute information, and association information in the BIM model, and ignores the multi-level characteristics of construction project data. Summary of the Invention

[0005] Aiming at the low clustering accuracy of high-dimensional data in construction projects based on BIM in the prior art, this application provides a visualization management method and system for construction projects based on BIM. By improving the DBSCAN clustering algorithm and performing clustering analysis according to the hierarchical relationship of the BIM model of the construction project, etc., the analysis accuracy is improved.

[0006] The purpose of this application is achieved through the following technical solutions.

[0007] One aspect of this application provides a visualization management method for construction projects based on BIM, including: S1, collecting construction project data to generate a data set containing geometric information, attribute information, and association information; S2, according to the hierarchical relationship of the BIM model of the construction project, dividing the data set into multiple hierarchical subsets; using an improved DBSCAN clustering algorithm to perform clustering analysis on each hierarchical subset; the improved DBSCAN clustering algorithm adjusts the distance metric in the clustering process according to the importance of the attributes by introducing an attribute weight mechanism; S3, combining the clustering results of different hierarchical subsets to obtain multiple data clusters of the data set; S4, classifying and archiving the data set according to the data clusters, and generating a semantically associated data collaboration array A according to the classification and archiving results; S5, performing dimensionality reduction processing on the semantically associated data collaboration array A to obtain a dimensionality-reduced data array; S6, constructing a spatial index of the dimensionality-reduced data array using a density-based spatial index method; using a distributed computing framework, dividing the BIM model of the construction project into multiple partitions, and performing distributed matching between the partitions and the spatial index to obtain a data matching point set; S7, performing visualization analysis according to the data matching point set to generate a management strategy.

[0008] Further, in S2, an improved DBSCAN clustering algorithm is used to perform clustering analysis on each hierarchical subset, including: S21, dividing the data set into multiple hierarchical subsets according to the hierarchical relationship of the building engineering BIM model; the hierarchical relationship includes buildings, floors, and components; S22, setting a dynamic weight matrix W, initializing the weight value, and defining the local data feature metric , the local data feature metric for measuring the local importance of the k-th attribute in the t-th iteration; S23, in each iteration, updating the dynamic weight matrix W according to the local data feature metric and the clustering result of the previous iteration; S24, calculating the weighted Euclidean distance and between any two data points ; S25, performing DBSCAN clustering on each hierarchical subset according to the weighted Euclidean distance , to obtain the data clustering result of the corresponding level; S26, repeating steps S23 to S25 until the clustering result converges or reaches the preset maximum number of iterations, to obtain the final clustering result of each level.

[0009] On the one hand, the traditional DBSCAN algorithm uses a single distance metric (such as Euclidean distance) to measure the similarity between data points, treating all data points as equal individuals. This processing method ignores the inherent hierarchical structure in building engineering BIM data, such as the relationships between different levels like buildings, floors, and components. The single distance metric cannot capture and represent the semantic differences and correlations between data at different levels, resulting in the clustering result being unable to accurately reflect the actual structure of the building project.

[0010] In this application, according to the hierarchical relationship of the building engineering BIM model, the data set is divided into multiple hierarchical subsets, such as buildings, floors, and components. This hierarchical division takes into account the multi-level characteristics of BIM data, enabling the clustering algorithm to better adapt to the structure and organization of building engineering data. Through independent clustering analysis of data at different levels, the internal correlations and patterns of data at each level can be discovered, improving the accuracy and interpretability of clustering.

[0011] In this application, when calculating the similarity between data points, the weighted Euclidean distance is adopted, combining the dynamic weight matrix with the attribute difference. By introducing attribute weights, the clustering algorithm can focus on the differences of important attributes when calculating distances, reducing the interference of secondary attributes.

[0012] Further, the weighted Euclidean distance and between any two data points is calculated through the following formula: , where represents the weighted distance at the t-th iteration ; represents the dynamic weight of the k-th attribute at the t-th iteration; and respectively represent the data points and 's values on the k-th attribute; k is the dimension of the attribute, and its value range is from 1 to n, where n is the total number of attributes in the dataset; t represents the number of clustering iterations, and its value range is from 1 to T, where T is the preset maximum number of iterations or the number of iterations when clustering converges; represents the local importance measure of the k-th attribute at the t-th iteration, which can be measured by calculating indicators such as variance and entropy of the attribute within the local neighborhood; , where α is the learning rate parameter, used to control the adjustment speed of the dynamic weight, and its value range is from 0 to 1.

[0013] , where represents the occurrence probability of the j-th value of the k-th attribute within the local neighborhood of the data point ; m represents the number of different values of the k-th attribute within the local neighborhood of the data point .

[0014] When the traditional DBSCAN algorithm processes building engineering BIM data, it usually adopts a static weight assignment method, that is, a fixed weight value is assigned to each attribute. However, this static weight assignment may not be able to adapt to the local characteristics and changes of building engineering data at different levels and regions. Different attributes may have different importance and influence at different levels and regions, and the static weight assignment ignores this dynamic change, resulting in limitations and inaccuracies in the clustering results.

[0015] This application introduces a dynamic weight matrix W and local data feature metrics to measure the importance of different attributes in the clustering process. By updating the dynamic weight matrix according to the local data feature metrics and the previous clustering results in each iteration, the clustering algorithm can adaptively adjust the attribute weights, fully considering the influence of different attributes on the clustering results. This dynamic weight adjustment mechanism can better capture the local patterns and changes of building engineering data at different levels and regions, making the clustering results more accurate and reliable.

[0016] Further, in S4, the data set is classified and archived according to data clustering, and a semantically associated data collaboration array A is generated according to the classification and archiving results, including: S41, extracting the common features of the data points within each data cluster as the semantic description of the corresponding cluster; the common features represent the consistency or similarity of the attribute distribution of the data points within the cluster; S42, classifying and archiving the data clusters according to the semantic description to obtain a data classification with semantic association; S43, traversing each data point in the data set and mapping the data point to the corresponding data classification according to the data cluster and semantic description to which the data point belongs; S44, extracting the attribute information of the data points within each data classification to construct an attribute matrix, where each row of the attribute matrix represents a data point and each column represents an attribute; the attribute includes geometric attributes; S45, constructing an association matrix according to the topological relationship and logical relationship between the data points, and the association matrix represents the association strength between the data points; S46, using a multi-dimensional array to store TensorFlow and combining the attribute matrix and the association matrix into a multi-dimensional data collaboration array.

[0017] Further, in S5, dimensionality reduction processing is performed on the semantically associated data collaboration array A to obtain a dimensionality-reduced data array, including: S51, performing centering processing on the semantically associated data collaboration array A to obtain a centered matrix ; The purpose of centering processing is to shift the mean of the data to the origin so that the mean of the data in each dimension is 0. This can eliminate the translational influence of the data and make the subsequent eigenvalue decomposition and dimensionality reduction processing more accurate and stable.

[0018] S52, introducing a kernel function , calculating the kernel matrix K, where, , represents the matrix data point and in the high-dimensional feature space; The kernel function can map the original data to a high-dimensional feature space and capture the non-linear features of the data. By calculating the inner product between data points in the high-dimensional feature space, the kernel matrix K is obtained. Here, the data points x_i and x_j are extracted from the centered matrix A'.

[0019] S53, performing centering processing on the kernel matrix K to obtain a centered kernel matrix ; S54, performing eigenvalue decomposition on the centered kernel matrix to obtain an eigenvalue matrix Λ and an eigenvector matrix V, that is, ; Eigenvalue decomposition can extract the main features and directions of the kernel matrix and provide a basis for subsequent dimensionality reduction processing.

[0020] S55, performing dimensionality reduction processing on the eigenvalue matrix Λ, selecting the top k largest eigenvalues to obtain a dimensionality-reduced eigenvalue matrix , where k is the dimension after dimensionality reduction; by selecting the most important eigenvalues and eigenvectors, the data is mapped from a high-dimensional space to a low-dimensional space.

[0021] S56. According to the eigenvalue matrix after dimensionality reduction , truncate the eigenvector matrix V to obtain the eigenvector matrix after dimensionality reduction ; S57. Calculate the co-occurrence matrix of the data after dimensionality reduction , that is , and use the co-occurrence matrix of the data after dimensionality reduction as the dimensionality-reduced data matrix. can be regarded as an approximate representation of the original data matrix A in the low-dimensional space.

[0022] Traditional Singular Value Decomposition (SVD) assumes a linear relationship between data points. However, the co-occurrence matrix A of semantically related data may have non-linear characteristics, and linear dimensionality reduction methods may not be able to effectively capture the internal structure and complex relationships of the data, resulting in the loss of important information in the dimensionality-reduced data matrix.

[0023] In this application, by introducing a kernel function, the original data is mapped to a high-dimensional feature space, and eigenvalue decomposition is performed in the high-dimensional feature space, which can capture the non-linear structure and complex relationships of the data. The choice of the kernel function can be determined according to the characteristics of the data and prior knowledge. Commonly used kernel functions include Gaussian kernel function and polynomial kernel function, etc. Perform eigenvalue decomposition on the centered kernel matrix to obtain the eigenvalue matrix Λ and the eigenvector matrix V. This is equivalent to performing Principal Component Analysis (PCA) in the high-dimensional feature space to extract the main features of the data. By introducing the kernel trick, feature extraction and dimensionality reduction can be performed in the high-dimensional feature space, effectively dealing with the non-linear characteristics in the co-occurrence matrix of semantically related data, and improving the effect of dimensionality reduction and the representation ability of the data.

[0024] Further, in S53, perform centering processing on the kernel matrix K to obtain the centered kernel matrix , including: calculating the centering matrix H: Calculate the centering matrix H, , where I is the n×n identity matrix, is the n×n all-ones matrix, and n is the number of data points; use the centering matrix H to perform centering processing on the kernel matrix K to obtain the centered kernel matrix : .

[0025] The centered kernel matrix K' makes the row and column means of the kernel matrix both 0, eliminating the bias term of the data. The centering processing can improve the effect of kernel principal component analysis, making the extracted features more accurate and stable. On the other hand, it avoids explicitly calculating the distances between data points in the original space, reducing the computational complexity.

[0026] The purpose of centering processing is to convert the row and column means of the kernel matrix to 0, eliminate the bias term of the data, and make the kernel matrix more suitable for eigenvalue decomposition. By introducing the centering matrix H, data centering can be achieved in the kernel space.

[0027] Furthermore, in S6, a spatial index of the dimensionality-reduced data array is constructed using a density-based spatial indexing method; using a distributed computing framework, the BIM model of the construction project is divided into multiple partitions, and the partitions are distributedly matched with the spatial index to obtain a data matching point set, including: S61, spatially partitioning the BIM model of the construction project into multiple partitions of equal size; S62, partitioning the dimensionality-reduced data array into multiple grid cells of equal data space size; S63, calculating the number of data points in each grid cell, and estimating the density of the grid cell based on the number of data points; S64, selecting the grid cells with density greater than the threshold as dense cells, and inserting the data points in the dense cells into the spatial index structure, where the spatial index structure uses an R-tree structure; S65, for the grid cells with density less than the threshold, recursively partitioning them until the density of all grid cells is greater than the threshold or the preset maximum partitioning depth is reached; S66, storing the data point information in the grid cell corresponding to each node in the spatial index structure, where the data point information includes the coordinates and attributes of the data points; S67, using the distributed computing framework, matching the partitions of the BIM model of the construction project with the corresponding spatial index structure to obtain the data matching point set within the partitions.

[0028] Another aspect of the present application further provides a BIM-based visual management system for construction projects, which is used to execute a BIM-based visual management method for construction projects of the present application.

[0029] Compared with the prior art, the advantages of the present application are as follows:

[0030] BIM models usually contain a large amount of multi-dimensional data such as geometry, materials, and attributes. Traditional clustering algorithms (such as K-means) are prone to falling into the "curse of dimensionality" when dealing with high-dimensional data, resulting in poor clustering effects and high computational complexity. The present application adopts the density-based clustering algorithm DBSCAN. First, according to the hierarchical relationship of the model, the data is divided into several hierarchical subsets (such as floors, specialties, component types, etc.). Then, DBSCAN clustering is independently performed within each subset. Finally, the clustering results of different levels are merged and coordinated to obtain the final clustering result. Hierarchical clustering can reduce the data complexity and improve the clustering efficiency and accuracy.

[0031] The traditional method of traversing the BIM data of the target construction project and matching it with the dimensionality-reduced data array means that a large number of similarity calculation and comparison operations are required. When the scale of the BIM model and the data array is large, this exhaustive matching process can be very time-consuming and inefficient. Especially in real-time application scenarios, such as construction progress tracking and design change analysis, the high computational cost may become a performance bottleneck, affecting the system's response speed and user experience. To accelerate the matching process in this application, an efficient index structure, such as a KD-tree, an R-tree, etc., is established for the dimensionality-reduced data array. The index is used to quickly lock potential matching candidates and avoid unnecessary similarity calculations. At the same time, parallel computing technology is adopted to distribute the matching tasks to multiple processing units, making full use of the computing resources and improving the matching efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] This application will be further described by way of exemplary embodiments, which will be described in detail through the drawings. These embodiments are not restrictive. In these embodiments, the same numbers represent the same structures, where:

[0033] Figure 1 is an exemplary flowchart of a BIM-based visualization management method for construction projects according to some embodiments of this application;

[0034] Figure 2 is an exemplary flowchart of obtaining the final clustering result according to some embodiments of this application;

[0035] Figure 3 is an exemplary flowchart of constructing a data collaboration array according to some embodiments of this application;

[0036] Figure 4 is an exemplary flowchart of generating a dimensionality-reduced data collaboration array according to some embodiments of this application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0037] The methods and systems provided by the embodiments of this application will be described in detail below with reference to the drawings.

[0038] As Figure 1As shown, collect construction project data to generate a dataset containing geometric information, attribute information, and association information; divide the dataset into multiple hierarchical subsets according to the hierarchical relationship of the construction project BIM model; perform clustering analysis on each hierarchical subset using an improved DBSCAN clustering algorithm; the improved DBSCAN clustering algorithm adjusts the distance metric in the clustering process by introducing an attribute weight mechanism according to the importance of attributes; merge the clustering results of different hierarchical subsets to obtain multiple data clusters of the dataset; classify and archive the dataset according to the data clusters, and generate a semantically associated data collaboration array A according to the classification and archiving results; perform dimensionality reduction processing on the semantically associated data collaboration array A to obtain a dimensionality-reduced data array; construct a spatial index for the dimensionality-reduced data array using a density-based spatial indexing method; use a distributed computing framework to divide the construction project BIM model into multiple partitions, and perform distributed matching between the partitions and the spatial index to obtain a data matching point set; perform visual analysis according to the data matching point set to generate a management strategy.

[0039] S1. Collect construction project data to generate a dataset containing geometric information, attribute information, and association information; model the construction project through BIM software (such as Autodesk Revit, Bentley AECOsim Building Designer, etc.) or other 3D modeling tools to obtain the geometric information of building components, including position coordinates, dimensions, shapes, etc. Extract the attribute information of building components, such as material, construction date, cost, fire resistance rating, etc., and associate it with the geometric information. Identify and record the association relationships between building components, such as the logical composition, topological relationship, and spatial relationship of components. Integrate the collected geometric information, attribute information, and association information into a structured dataset, such as XML, JSON, etc. format, for subsequent data processing and analysis.

[0040] As Figure 2 shown, S2. Perform clustering analysis on each hierarchical subset using an improved DBSCAN clustering algorithm, including: S21. Divide the dataset into multiple hierarchical subsets according to the hierarchical relationship of the construction project BIM model; the hierarchical relationship includes buildings, floors, and components; analyze the hierarchical structure of the BIM model to identify the three main levels of buildings, floors, and components. According to the hierarchical relationship, divide the data points in the dataset into subsets at the three levels of buildings, floors, and components. For each hierarchical subset, extract relevant geometric information, attribute information, and association information to form an independent data subset.

[0041] S22. Set a dynamic weight matrix W, initialize the weight values, and define local data feature metrics : Create an n×T dynamic weight matrix W, where n is the total number of attributes in the dataset and T is the maximum number of clustering iterations. Initialize the first column of the weight matrix W, that is The initial weight values can be set according to prior knowledge or statistical analysis. For example, the initial weights of all attributes can be set to 1 / n, indicating that the importance of each attribute is the same initially. Define the local data feature metric to measure the local importance of the k-th attribute in the t-th iteration. Different metric indicators can be selected, such as variance, entropy, etc.

[0042] S23. In each iteration, according to the local data feature metric and the clustering result of the previous iteration, update the dynamic weight matrix . Taking the data point as the center, select the k nearest data points as the local neighborhood . The value of k can also be set according to the characteristics of the dataset and prior knowledge, or optimized by methods such as cross-validation. In the local neighborhood Ni, extract the values of the k-th attribute. For each data point in the local neighborhood , obtain the value of its k-th attribute, forming a local attribute value set. Calculate the number m of different values of the k-th attribute in the local neighborhood . Traverse the local attribute value set and count the number of different values to obtain the value of m. Calculate the occurrence probability of each value of the k-th attribute in the local neighborhood . For each value in the local attribute value set, count its occurrence frequency and divide by the total number of data points in the local neighborhood to obtain the occurrence probability of this value.

[0043] Calculate the local data feature metric : , where represents the occurrence probability of the j-th value of the k-th attribute in the local neighborhood of the data point; m represents the number of different values of the k-th attribute in the local neighborhood of the data point . According to the local data feature metric

[0044] , update the t-th column of the dynamic weight matrix W, that is . Use the exponential decay function for weight update: , where , represents the dynamic weight of the k-th attribute in the t-th iteration; α is the learning rate parameter that controls the speed of weight adjustment. Repeat until the local data feature metrics of all attributes are calculated and the t-th column of the dynamic weight matrix W is updated.

[0045] S24, Calculate the weighted Euclidean distance between any two data points and : For any two data points in the dataset and and , extract their values on each attribute, denoted as and respectively, where k is the dimension of the attribute, and the value range is from 1 to n. According to the current iteration number t, obtain the corresponding attribute weight vector from the dynamic weight matrix W, that is . Calculate the weighted Euclidean distance , and the formula is as follows: , where represents the dynamic weight of the k-th attribute in the t-th iteration; represents the difference between the data points and on the k-th attribute. Return the calculated weighted Euclidean distance , which is used for the subsequent clustering process. In each iteration, the attribute weights are dynamically adjusted according to the local data characteristics measurement, so that the clustering algorithm can adaptively consider the importance of different attributes in the local area. This dynamic weight mechanism helps to improve the accuracy and interpretability of the clustering results, and better discover the local patterns and associations in the construction engineering data. At the same time, through the calculation of the weighted Euclidean distance, the attribute weights and the differences between data points can be comprehensively considered, improving the adaptability and robustness of the clustering algorithm.

[0046] S25, According to the weighted Euclidean distance , perform DBSCAN clustering on each hierarchical subset to obtain the data clustering results at the corresponding level; for each hierarchical subset, set the parameters of the DBSCAN algorithm, including the distance threshold ε and the minimum point number threshold MinPts. These parameters can be set according to the characteristics of the hierarchical subset and prior knowledge, or optimized through methods such as cross-validation.

[0047] For each data point in the hierarchical subset, calculate its weighted Euclidean distance from other data points , using the distance calculation formula in step S24. According to the distance threshold ε, determine the ε-neighborhood of each data point. For the data point , its ε-neighborhood includes all data points whose weighted Euclidean distance from does not exceed ε.

[0048] According to the minimum point number threshold MinPts, determine the type of each data point: Core point: If the number of data points within the ε-neighborhood of the data point is greater than or equal to MinPts, then Marked as core points. Border points: If the number of data points within the ε-neighborhood of a data point is less than MinPts, but is within the ε-neighborhood of a certain core point, then is marked as a border point. c. Noise points: If a data point is neither a core point nor a border point, then is marked as a noise point.

[0049] For each core point, perform a density-reachable clustering process: Select an unvisited core point as the starting point of the current cluster and create a new cluster. Add all directly density-reachable points of (i.e., points within the ε-neighborhood of ) to the current cluster and mark them as visited. For each point in the current cluster, recursively add its directly density-reachable points to the cluster until no new points can be added. Repeat until all core points are assigned to clusters.

[0050] Assign border points to the clusters of the core points they belong to. For each border point, find the core point within its ε-neighborhood and assign the border point to the cluster of that core point. If there are multiple core points within the ε-neighborhood of a border point, the cluster of the core point closest in distance can be selected. Mark noise points as separate clusters. Return the clustering results of the hierarchical subsets, including information such as the labels of the clusters, the data points they belong to, etc. S26. Repeat steps S23 to S25 until the clustering results converge or reach the preset maximum number of iterations to obtain the final clustering results for each level;

[0051] S3. Merge the clustering results of different hierarchical subsets to obtain multiple data clusters for the dataset; According to the association relationships between levels, merge and integrate the clustering results of different levels: For the clusters at the building level, associate them with the corresponding floor-level clusters to form a building - floor hierarchical structure. For the clusters at the floor level, associate them with the corresponding component-level clusters to form a floor - component hierarchical structure. For the clusters at the component level, analyze the floors and buildings they belong to to form a complete building - floor - component hierarchical structure. Generate the final data clustering results, including information such as the labels of the clusters, the levels they belong to, the data points they contain, the cluster centers, etc.

[0052] Such as Figure 3As shown in S4, classify and file the data set according to data clustering, and generate a semantically associated data collaboration array A according to the classification and filing results, including: S41, extract the common features of the data points within each data cluster as the semantic description of the corresponding cluster; the common features represent the consistency or similarity of the attribute distribution of the data points within the cluster: for each data cluster, analyze the attribute distribution of the building component data points within it. Extract the common geometric attributes of the components within the cluster, such as length, width, height, volume, etc., calculate statistical indicators such as their mean, median, mode, etc., as the geometric features of the cluster. Extract the common material attributes of the components within the cluster, such as material type, color, transparency, etc., and statistically analyze their distribution, and identify the materials with higher frequencies of occurrence as the material features of the cluster. Extract the common functional attributes of the components within the cluster, such as the room type, structural type, equipment type, etc. to which they belong, and statistically analyze their distribution, and identify the functional attributes with higher frequencies of occurrence as the functional features of the cluster. Considering geometric features, material features, and functional features comprehensively, generate semantic descriptions that can reflect the consistency or similarity of the attributes of the components within the cluster, such as "rectangular steel columns" and "glass curtain wall panels".

[0053] S42, classify and file the data clusters according to the semantic descriptions to obtain data classifications with semantic associations; analyze the semantic descriptions of each data cluster, extract the keywords and attribute labels therein, such as "column", "beam", "wall", "door", "window", etc. Group and classify the data clusters according to the similarity of the keywords and attribute labels. Clusters with similar semantic descriptions are grouped into the same data classification, such as "structural components", "enclosure components", "equipment components", etc. Assign a unique identifier or name to each data classification, such as "S_column", "S_beam", "A_wall", "A_door", "M_equipment", etc., for subsequent reference and association. According to the professional division and design stage of the construction project, hierarchically organize the data classifications to generate a tree structure or atlas of the classifications, representing the subordinate and associated relationships between the classifications.

[0054] S43, traverse each data point in the data set, and map the data point to the corresponding data classification according to the data cluster and semantic description to which the data point belongs; for each BIM component data point in the construction project, determine the data cluster to which it belongs. Obtain the semantic description and the corresponding data classification identifier of this data cluster. Map the BIM component data point to the corresponding data classification, and establish the association relationship between the component and the classification. Record the position or index of the BIM component in the data classification for subsequent retrieval and access. Repeat until all BIM component data points are traversed to complete the mapping process from the component to the data classification.

[0055] S44. Extract the attribute information of the data points within each data classification to construct an attribute matrix. Each row of the attribute matrix represents a data point, and each column represents an attribute. The attributes include geometric attributes. For each data classification, extract the attribute information of the BIM component data points within it. According to the attribute types and quantities of the BIM components, construct the structure of the attribute matrix. The number of rows of the matrix corresponds to the number of components, and the number of columns corresponds to the number of attributes. For each BIM component, extract the values of its geometric attributes (such as length, width, height, volume, etc.) and other relevant attributes (such as material, function, floor it belongs to, etc.), and fill them into the corresponding rows of the attribute matrix. For missing attribute values, default values, interpolation, or special marks can be used for filling to maintain the integrity of the attribute matrix. Repeat until the attribute information of all BIM components is filled to form a complete attribute matrix.

[0056] S45. According to the topological and logical relationships between the data points, construct an association matrix, which represents the association strength between the data points. Analyze the topological relationships between BIM components, such as adjacent relationships, connection relationships, inclusion relationships, etc., to determine the existing spatial associations between the components. Analyze the logical relationships between BIM components, such as subordination relationships, bearing relationships, transmission relationships, etc., to determine the existing semantic associations between the components. According to the types and strengths of the topological and logical relationships, assign an association strength value to each pair of BIM components to represent the closeness of their association. For example, the association strength between an adjacent wall and a column is higher than that between components that are far apart. Construct the structure of the association matrix. The number of rows and columns of the matrix both correspond to the number of BIM components. The elements of the matrix represent the association strength between the components. Traverse each pair of BIM components and fill the association strength value between them into the corresponding position of the association matrix. For component pairs with no direct association, 0 or special marks can be used in the association matrix to represent. Complete the construction of the association matrix, which represents the association strength and connection relationship between BIM components.

[0057] S46. Store TensorFlow using a multi-dimensional array, and combine the attribute matrix and the correlation matrix into a multi-dimensional data collaboration array. Convert the attribute matrix and the correlation matrix into a multi-dimensional array format supported by TensorFlow, such as tf.Tensor. Determine the structure and dimensions of the multi-dimensional data collaboration array according to the dimensions and shapes of the attribute matrix and the correlation matrix. For example, the attribute matrix can be used as the first dimension and the correlation matrix as the second dimension. Use TensorFlow's multi-dimensional array operations and functions, such as tf.concat, tf.stack, etc., to combine the attribute matrix and the correlation matrix into a unified multi-dimensional data collaboration array. Perform necessary dimension adjustments and transformations on the multi-dimensional data collaboration array, such as transposing, reshaping, etc., to meet the needs of subsequent data analysis and model training. Store the multi-dimensional data collaboration array in a TensorFlow data structure, such as tf.Variable or tf.constant, for subsequent access and calculation.

[0058] As Figure 4 shown, S5. Perform dimensionality reduction on the semantic correlation data collaboration array A to obtain a dimensionality-reduced data array, including: S51. Center the semantic correlation data collaboration array A to obtain a centered matrix ; calculate the mean of each attribute column of the data collaboration array A to obtain a mean vector m. For each element of the data collaboration array A, subtract the mean of the corresponding attribute column to obtain a centered element , that is . Form a new matrix from the centered elements as the centered data collaboration array.

[0059] S52. Introduce a kernel function , and calculate the kernel matrix K, where represents the inner product of data points and in a high-dimensional feature space; select a kernel function suitable for the characteristics of building engineering BIM data, such as a Gaussian kernel function, a polynomial kernel function, etc. For each pair of data points and in the data collaboration array A, calculate their inner product in the high-dimensional feature space to obtain the element of the kernel matrix K. Repeat until the kernel function values of all data point pairs are calculated to obtain a complete kernel matrix K.

[0060] S53. Center the kernel matrix K to obtain a centered kernel matrix ; calculate the centering matrix H, , where I is an n×n identity matrix, is an all - ones matrix of n×n, where n is the number of data points; according to the centering matrix H, the centering process of the kernel matrix K is carried out through the following formula: .

[0061] S54, perform eigenvalue decomposition on the centered kernel matrix to obtain the eigenvalue matrix Λ and the eigenvector matrix V, that is ; perform eigenvalue decomposition on the centered kernel matrix to obtain the eigenvalue matrix Λ and the eigenvector matrix V. Arrange the eigenvalues in descending order, and arrange the corresponding eigenvectors in the same order.

[0062] S55, perform dimensionality reduction on the eigenvalue matrix Λ, select the first k largest eigenvalues to obtain the dimensionality - reduced eigenvalue matrix , where k is the dimensionality after dimensionality reduction; according to the characteristics and analysis requirements of BIM component data, determine the target dimensionality k after dimensionality reduction. Select the first k largest eigenvalues in the eigenvalue matrix Λ to obtain the dimensionality - reduced eigenvalue matrix . For example, in the construction schedule and resource optimization of construction projects, the BIM model contains a large amount of data related to construction activities and resource allocation, such as task duration, dependencies, resource requirements, etc. By performing eigenvalue decomposition on this data, the construction schedule and resource allocation can be optimized. According to the scale and complexity of the construction project, the first 4 - 8 largest eigenvalues can be selected, and these eigenvalues usually can reflect the main bottlenecks and influencing factors of the construction schedule and resource allocation. Select the first k largest eigenvalues in the eigenvalue matrix Λ to form the dimensionality - reduced eigenvalue matrix , where k is 4 - 8. S56, according to the dimensionality - reduced eigenvalue matrix , truncate the eigenvector matrix V to obtain the dimensionality - reduced eigenvector matrix ; S57, calculate the dimensionality - reduced data collaboration array , that is , and take the dimensionality - reduced data collaboration array as the dimensionality - reduced data array.

[0063] S6. Construct a spatial index for the dimensionality-reduced data array using a density-based spatial indexing method; using a distributed computing framework, divide the building engineering BIM model into multiple partitions, and perform distributed matching between the partitions and the spatial index to obtain a data matching point set, including: S61. Perform spatial partitioning on the building engineering BIM model, dividing it into multiple partitions of equal size; determine the size and number of partitions according to the spatial range and analysis requirements of the building engineering BIM model. Use methods such as uniform grid partitioning or octree partitioning to divide the BIM model into multiple partitions of equal size. Assign a unique identifier to each partition for subsequent distributed processing and index matching.

[0064] S62. Divide the dimensionality-reduced data array into multiple grid cells of equal data space size; determine the size and number of grid cells according to the spatial range and analysis requirements of the dimensionality-reduced data array. Use the method of uniform grid partitioning to divide the dimensionality-reduced data array into multiple grid cells of equal size. Assign a unique identifier to each grid cell for subsequent density estimation and index construction.

[0065] S63. Calculate the number of data points in each grid cell, and estimate the density of the grid cell based on the number of data points; for the dimensionality-reduced data array of the building engineering BIM model, divide it into multiple grid cells of equal size. Initialize a hash table or dictionary structure to store the number of data points in each grid cell. Traverse each data point in the dimensionality-reduced data array: calculate the grid cell index to which the data point belongs, which can be calculated based on the coordinates of the data point and the size of the grid cell. Look up the corresponding grid cell index in the hash table or dictionary. If it does not exist, add it to the hash table or dictionary and initialize the number of data points to 1; if it already exists, increment the number of data points by 1. Traverse each grid cell in the hash table or dictionary: obtain the number of data points in the grid cell. Calculate the volume or area of the grid cell (select according to the actual situation). Calculate the data point density of the grid cell, that is, the number of data points divided by the volume or area of the grid cell. Store the density estimation value of the grid cell in a new array or list, with the index corresponding to the index of the grid cell. Return the array or list of density estimation values of the grid cells.

[0066] S64. Select grid cells with a density greater than the threshold as dense cells, and insert the data points within the dense cells into the spatial index structure, where the spatial index structure adopts an R-tree structure; set a density threshold, which is determined according to the characteristics and analysis requirements of the building engineering BIM data. The appropriate threshold can be determined through empirical values, statistical analysis, or parameter tuning. Initialize an R-tree index structure for storing the data points within the dense cells. Traverse the array or list of density estimates of the grid cells: If the density of a grid cell is greater than the set threshold, mark it as a dense cell. For each dense cell, traverse the data points within it: Package the coordinate and attribute information of the data point into a data object. Insert the data object into the R-tree index structure, and perform spatial partitioning and organization according to the coordinates of the data points. After inserting all the data points within the dense cells, return the constructed R-tree index structure.

[0067] S65. For grid cells with a density less than the threshold, recursively divide them until the density of all grid cells is greater than the threshold or the preset maximum division depth is reached; for grid cells with a density less than the threshold, divide them into smaller sub-grid cells. The division method can be selected as needed, such as quadtree division or octree division. For each sub-grid cell, recursively perform the following steps: Calculate the data point density of the sub-grid cell in a similar way to S63. If the density of the sub-grid cell is greater than the threshold, mark it as a dense cell and insert the data points within it into the R-tree index structure in a similar way to S64. If the density of the sub-grid cell is less than the threshold and the current division depth has not reached the preset maximum division depth, recursively divide the sub-grid cell. If the density of the sub-grid cell is less than the threshold and the current division depth has reached the preset maximum division depth, mark it as a sparse cell and do not perform further division. After completing the recursive division of all grid cells, return the updated R-tree index structure.

[0068] S66. On each node of the spatial index structure, store the data point information within the grid cell corresponding to the node, where the data point information includes the coordinates and attributes of the data points; on each node of the R-tree index structure, in addition to storing the MBR information of the node, it is also necessary to store the data point information within the grid cell corresponding to the node. For each data point, extract its coordinate information (such as three-dimensional coordinates) and attribute information (such as component type, material, size, etc.). Package the coordinate and attribute information of the data point into a data object and associate it with the corresponding R-tree node. Linked lists, arrays, or other data structures can be used to store the data point information within the node for quick access and retrieval. On the leaf nodes of the R-tree index structure, store the actual data point information; on the non-leaf nodes, store the summary or statistical information of the data point information contained in its child nodes, such as the number of data points, coordinate range, etc.

[0069] S67. Using a distributed computing framework, match the partitions of the building engineering BIM model with the corresponding spatial index structures to obtain the data matching point sets within the partitions; use a distributed computing framework (such as Apache Spark, Hadoop, etc.) to distribute the partitions of the building engineering BIM model to multiple computing nodes. On each computing node, load the BIM model data and spatial index structures corresponding to the partition. For each partition, traverse the BIM components within it, quickly retrieve through the spatial index structure, and find the data points that match the spatial positions of the components. Associate the matching data points with the corresponding BIM components to form the data matching point sets within the partitions. Aggregate and merge the data matching point sets of each partition to obtain the data matching point set of the entire building engineering BIM model.

[0070] S7. Conduct visual analysis based on the data matching point sets to generate management strategies. By analyzing the matching data point sets, the implementation of the construction progress can be tracked and visualized in real time. Compare the actual construction progress with the planned progress to identify tasks or work areas with progress deviations and lags. According to the progress deviation situation, generate warning information to prompt project managers to pay attention to and handle potential progress risks. By visually displaying the severity and scope of the progress deviation, help managers prioritize the handling of key issues. Also, by analyzing the matching data point sets, the resource requirements for a period of time in the future, such as labor, materials, equipment, etc., can be predicted. Combining the construction progress plan and the actual progress situation, dynamically adjust the resource requirement prediction to ensure that the resource supply matches the actual demand. Based on the resource requirement prediction and the real-time available resource situation, generate resource optimization scheduling strategies to reasonably allocate and schedule resources. By visually displaying the allocation and usage of resources, identify resource bottlenecks and idleness, and support dynamic adjustment and optimization.

Claims

1. A BIM-based construction project visualization management method, characterized in that: include: S1, collects construction engineering data and generates a data set containing geometric information, attribute information and association information; S2, according to the hierarchical relationship of the BIM model of the construction project, the data set is divided into multiple hierarchical subsets; An improved DBSCAN clustering algorithm is used to perform cluster analysis on each hierarchical subset; the improved DBSCAN clustering algorithm adjusts the distance metric in the clustering process according to the importance of the attribute by introducing an attribute weight mechanism; S3, merging the clustering results of subsets at different levels to obtain multiple data clusters of the data set; S4, classify and archive the data set according to data clustering, and generate a semantically associated data collaborative array A according to the classification and archiving results; S5, performing dimensionality reduction processing on the semantically associated data coordination array A to obtain a dimensionality-reduced data array; S6, constructing a spatial index of the reduced-dimensional data array using a density-based spatial indexing method; Using the distributed computing framework, the BIM model of the construction project is divided into multiple partitions, and the partitions are distributedly matched with the spatial index to obtain a data matching point set; S7, perform visual analysis based on the data matching point set and generate a management strategy.

2. The BIM-based construction project visualization management method according to claim 1 is characterized by: S2, using the improved DBSCAN clustering algorithm to perform cluster analysis on each level subset, including: S21, dividing the data set into a plurality of hierarchical subsets according to the hierarchical relationship of the building engineering BIM model; the hierarchical relationship includes buildings, floors, and components; S22, set the dynamic weight matrix W, initialize the weight value, and define the local data feature measurement , local data feature measurement Used to measure the local importance of the k-th attribute in the t-th iteration; S23, in each iteration, according to the local data feature measurement And the clustering result of the previous iteration, update the dynamic weight matrix W; S24, calculate any two data points and The weighted Euclidean distance between ; S25, based on weighted Euclidean distance , perform DBSCAN clustering on each level subset to obtain the data clustering results of the corresponding level; S26, repeating steps S23 to S25 until the clustering results converge or the preset maximum number of iterations is reached, to obtain the final clustering results of each level.

3. The BIM-based construction project visualization management method according to claim 2 is characterized by: Calculate the value of any two data points and The weighted Euclidean distance between , through the following formula: ； in, represents the weighted distance of the tth iteration ; represents the dynamic weight of the k-th attribute in the t-th iteration; and Represents data points and The value of the kth attribute; k is the dimension of the attribute, ranging from 1 to n; t represents the number of clustering iterations.

4. The BIM-based construction project visualization management method according to claim 3 is characterized by: ； Among them, α is the learning rate parameter; represents the local importance measure of the k-th attribute in the t-th iteration.

5. The BIM-based construction project visualization management method according to claim 4 is characterized by: ； in, Indicates that the kth attribute is in the data point Local area The probability of occurrence of the jth value in the data point; m represents the probability of occurrence of the kth attribute in the data point Local area The number of different values ​​in .

6. The BIM-based construction project visualization management method according to claim 1, characterized in that: S4, classifying and archiving the data set according to data clustering, and generating a semantically associated data coordination array A according to the classification and archiving results, including: S41, extracting common features of data points within each data cluster as a semantic description of the corresponding cluster; the common features represent the consistency or similarity of attribute distribution of data points within the cluster; S42, classifying and archiving the data clusters according to the semantic description to obtain data classifications with semantic associations; S43, traversing each data point in the data set, and mapping the data point to a corresponding data classification according to the data cluster and semantic description to which the data point belongs; S44, extracting attribute information of data points within each data classification, and constructing an attribute matrix, wherein each row of the attribute matrix represents a data point, and each column represents an attribute; the attributes include geometric attributes; S45, constructing a correlation matrix according to the topological relationship and logical relationship between the data points, wherein the correlation matrix represents the correlation strength between the data points; S46, uses multidimensional arrays to store TensorFlow and combines the attribute matrix and the association matrix into a multidimensional data collaborative array.

7. The BIM-based construction project visualization management method according to claim 6 is characterized by: S5, performing dimensionality reduction processing on the semantically associated data coordination array A to obtain a dimensionality-reduced data array, including: S51, centralize the semantically associated data collaborative array A to obtain a centralized matrix ; S52, introduce kernel function , calculate the kernel matrix K, where , which represents the matrix Medium data point and Inner products in high-dimensional feature spaces; S53, centralize the kernel matrix K to obtain the centralized kernel matrix ; S54, the centralised kernel matrix Perform eigenvalue decomposition to obtain the eigenvalue matrix Λ and the eigenvector matrix V, that is, ; S55, perform dimensionality reduction processing on the eigenvalue matrix Λ, select the first k largest eigenvalues, and obtain the eigenvalue matrix after dimensionality reduction , where k is the dimension after dimensionality reduction; S56, according to the eigenvalue matrix after dimensionality reduction , truncate the eigenvector matrix V, and get the eigenvector matrix after dimensionality reduction ; S57, calculate the data coordination array after dimensionality reduction ,Right now , the reduced-dimensional data is coordinated into an array as a dimensionally reduced data array.

8. The BIM-based construction project visualization management method according to claim 7 is characterized by: S53, centralize the kernel matrix K to obtain the centralized kernel matrix ,include: Calculate the centering matrix H, , where I is the n×n identity matrix, is an n×n all-1 matrix, where n is the number of data points; According to the centralization matrix H, the kernel matrix K is centralized by the following formula: 。 9. The method for visual management of construction projects based on BIM according to any one of claims 2 to 8, characterized in that: S6, obtain the data matching point set, including: S61, spatially partitioning the construction engineering BIM model into a plurality of partitions of equal size; S62, dividing the dimension-reduced data array into a plurality of grid units of equal data space size; S63, calculating the number of data points in each grid cell, and estimating the density of the grid cell according to the number of data points; S64, selecting a grid cell with a density greater than a threshold as a dense cell, and inserting the data points in the dense cell into a spatial index structure, wherein the spatial index structure adopts an R-tree structure; S65, recursively dividing the grid cells whose density is less than the threshold value until the density of all grid cells is greater than the threshold value or a preset maximum division depth is reached; S66, storing data point information in a grid unit corresponding to the node at each node of the spatial index structure, where the data point information includes coordinates and attributes of the data point; S67, using a distributed computing framework, matching the partitions of the construction engineering BIM model with the corresponding spatial index structure to obtain a data matching point set within the partition.

10. A BIM-based construction project visualization management system, characterized in that: include: At least one processing unit; used to execute instructions to implement the BIM-based construction project visualization management method as described in any one of claims 1 to 9.

Images