BIM technology and engineering quantity intelligent verification method and system

Abstract

The invention provides a BIM technology and project amount intelligent verification method and system, and the method comprises the steps: carrying out the word segmentation and part-of-speech tagging of a progress log, extracting an excavation region, a completion amount and a timestamp of the day, and obtaining a construction progress information set; according to the construction progress information set, identifying a semantic relationship of an excavation area, a completion volume and a time triple, and determining time-space semantic representation; extracting geometric information of each target earthwork logic body from the set of earthwork logic bodies, and calculating a design engineering quantity of each associated logic body in a corresponding preset elevation range in real time in combination with corresponding elevation data extracted from the BIM model to generate a design quantity data set; and according to the early warning result, updating the state attribute of the earthwork logic body in the BIM model, generating a dynamically updated BIM model state in combination with the deviation occurrence timestamp and the difference value, extracting the spatial distribution of the earthwork logic body in the BIM model, and generating a construction path.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a BIM technology and an intelligent verification method and system for engineering quantities. Background Art

[0002] Building Information Modeling (BIM) technology is increasingly being used in the field of engineering construction. It uses digital means to manage the entire life cycle of a project, which is of key significance to improving project efficiency and quality. Especially in complex construction scenarios such as earthwork excavation, BIM can provide precise support for engineering quantity verification and progress management. However, existing methods still have significant limitations when processing on-site progress data. Many systems rely on manually organized structured data and cannot effectively deal with the unstructured or semi-structured logs commonly found on construction sites, making it difficult to quickly and accurately associate progress information with BIM models. In addition, dynamic deviation control and real-time early warning mechanisms based on geological complexity are often missing in existing solutions, which can easily lead to deviations in engineering quantity verification and affect decision-making efficiency. These limitations expose some technical challenges. How to automatically parse key information, such as excavation areas and completed volumes, from unstructured progress logs, and associate the parsed progress information with the predefined earthwork logical body in the BIM model through spatiotemporal semantic matching, and calculate the design engineering quantity of the corresponding elevation in real time. At the same time, dynamically adjust the deviation threshold according to the complexity of the adjacent geological drilling data, and implement a highlighted warning when the difference between the actual and designed quantities exceeds the standard, all test the intelligence and real-time performance of the system. These technical factors have not been resolved, resulting in low efficiency in progress verification and insufficient deviation control, which restricts the application of BIM in complex construction scenarios. Therefore, how to combine named entity recognition and relationship extraction technology to achieve automatic information parsing in the scenario of layered and block excavation of foundation pits, associate the earthwork logical body of the BIM model through spatiotemporal semantic matching, calculate the design engineering quantity in real time, and dynamically adjust the deviation threshold according to the geological complexity to highlight the warning, has become a key issue. Summary of the Invention

[0003] The present invention provides a BIM technology and engineering quantity intelligent verification method, which mainly includes:

[0004] Perform word segmentation and part-of-speech tagging on the progress log to extract the excavation area, completed volume, and timestamp for the day to obtain a collection of construction progress information;

[0005] Based on the construction progress information set, the semantic relationship between the excavation area, completed volume and time triples is identified to determine the spatiotemporal semantic representation;

[0006] Obtain the engineering design drawings and establish a BIM model containing a preset earthwork logical body. Combine the spatiotemporal semantic representation with the established BIM model, map the excavation area in the semantic relationship diagram with the earthwork logical body in the BIM model using a unique ID, associate the progress information with the model entity, and obtain the earthwork logical body.

[0007] From the set of earthwork logical volumes, the geometric information of each target earthwork logical volume is extracted, and combined with the corresponding elevation data extracted from the BIM model, the design engineering quantity of each associated logical volume within the corresponding preset elevation range is calculated in real time to generate a design quantity data set;

[0008] The geological complexity level is assessed based on the distribution characteristics of the geological complexity adjacent to the excavation area, and the allowable deviation is dynamically adjusted according to the complexity level. A dynamic deviation threshold set is calculated. When the difference between the cumulative or single reported actual completed volume data and the design volume data set exceeds the threshold of the corresponding area and complexity in the dynamic deviation threshold set, the warning condition is triggered, the corresponding earthwork logical body is marked in the BIM model, and the warning result is obtained;

[0009] According to the early warning results, the status attributes of the earthwork logical body in the BIM model are updated. Combined with the deviation timestamp and difference, a dynamically updated BIM model status is generated. The spatial distribution of the earthwork logical body in the BIM model is extracted to generate the construction path.

[0010] The present invention provides a BIM technology and engineering quantity intelligent verification system, which mainly includes:

[0011] The progress information processing module is used to perform word segmentation and part-of-speech tagging on the progress log, extract the excavation area, completed volume and timestamp of the day, and obtain the construction progress information set;

[0012] The semantic relationship recognition module is used to identify the semantic relationship between the excavation area, completed volume and time triples based on the construction progress information set, and determine the spatiotemporal semantic representation;

[0013] The model mapping module is used to obtain engineering design drawings and establish a BIM model containing a preset earthwork logical body. By combining the spatiotemporal semantic representation with the established BIM model, the excavation area in the semantic relationship diagram is mapped to the earthwork logical body in the BIM model using a unique ID, and the progress information is associated with the model entity to obtain the earthwork logical body.

[0014] The engineering quantity calculation module is used to extract the geometric information of each target earthwork logical body from the set of earthwork logical bodies, and combine it with the corresponding elevation data extracted from the BIM model to calculate the design engineering quantity of each associated logical body within the corresponding preset elevation range in real time to generate a design quantity data set;

[0015] The early warning analysis module is used to evaluate the geological complexity level based on the distribution characteristics of the geological complexity of the excavation area, and dynamically adjust the allowable deviation according to the complexity level. The dynamic deviation threshold set is calculated. When the difference between the accumulated or single reported actual completed volume data and the design volume data set exceeds the threshold of the corresponding area and complexity in the dynamic deviation threshold set, the early warning condition is triggered, the corresponding earthwork logical body is marked in the BIM model, and the early warning result is obtained;

[0016] The status update module is used to update the status attributes of the earthwork logical body in the BIM model according to the early warning results, generate a dynamically updated BIM model status based on the deviation occurrence timestamp and difference, extract the spatial distribution of the earthwork logical body in the BIM model, and generate a construction path.

[0017] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0018] The present invention discloses a BIM technology and an intelligent verification method for engineering quantities. The method extracts structured progress information by semantically analyzing the progress log and establishes an association with the earthwork logic body in the BIM model. Combined with the geological complexity assessment, the allowable deviation threshold is dynamically adjusted to achieve intelligent comparison and early warning of the actual completed volume and the designed volume. The present invention can also update the BIM model status based on the early warning results, analyze the deviation trend, predict future risks, and optimize the excavation plan. This method realizes accurate monitoring, intelligent early warning and dynamic optimization of the earthwork progress, improves the construction management efficiency and quality control level, and provides strong support for the refined management of large-scale earthwork projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 This is a flow chart of a BIM technology and engineering quantity intelligent verification method of the present invention.

[0020] Figure 2 This is a structural schematic diagram of a BIM technology and engineering quantity intelligent verification system of the present invention. DETAILED DESCRIPTION

[0021] To further understand the content of the present invention, the present invention is described in detail with reference to the accompanying drawings and examples. The present application is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the relevant invention and are not intended to limit the invention. It should also be noted that, for ease of description, only portions relevant to the invention are shown in the accompanying drawings.

[0022] It can be understood that the BIM model refers to a specific application form of Building Information Modeling. It is not a fixed, unified entity, but a digital representation that is dynamically constructed according to specific project requirements and data. Specifically, the BIM model is realized by obtaining engineering design drawings and establishing a three-dimensional model containing preset earthwork logical bodies. Each earthwork logical body is the result of block processing based on its depth value, area value and elevation value, and is given a unique identifier (such as the "area number-layer number-serial number" format). In this model, each earthwork logical body not only contains its geometric information, such as the three-dimensional coordinate value of the center point and the coordinate value of the boundary vertex.

[0023] like Figure 1-2 In this embodiment, a BIM technology and engineering quantity intelligent verification method may specifically include:

[0024] Step S101 : perform word segmentation and part-of-speech tagging on the progress log, extract the excavation area, completed volume and timestamp of the day, and obtain a set of construction progress information.

[0025] Keywords are extracted from the progress log text using a word frequency statistics method to establish a construction professional dictionary, which contains fields such as the name of the construction area, the value of the excavation project quantity, and the construction time. Based on the construction area information in the construction professional dictionary, the latitude and longitude coordinates and geological parameters of the construction area are obtained from a preset engineering geographic information database to obtain a construction area feature data set. For the construction area feature data set, the excavation depth and excavation area of the construction area are calculated to obtain construction quantitative index data. The random forest algorithm is used to analyze the construction quantitative index data, and the construction completion rate is calculated using the least squares method. The construction status classification results are divided according to the completion rate interval to form a construction progress information set.

[0026] For example, keywords are extracted from the collected progress log text using a word frequency statistics method. A construction-specific dictionary containing fields such as construction area name, excavation quantity, and construction time is established. Progress information is then extracted from the dictionary and converted into structured data records. For the construction area information in the structured data records, the latitude and longitude coordinates of the construction points are obtained from a pre-set engineering geographic information database and correlated with the geological parameters of the construction area to form a construction area feature dataset. Based on the construction area feature dataset, the excavation depth, area, and volume of each construction area are calculated using an earthwork excavation volume calculation function, where earthwork excavation volume = excavation depth × excavation area, to obtain construction quantitative indicator data. A random forest algorithm is used to establish a construction progress prediction model, using the construction quantitative indicator data as input features. The construction completion rate is calculated using the least squares method to obtain a progress forecast for each construction area. Based on the progress forecast values, quantitative progress assessment rules are set. The construction area status is classified into three states: not started, in progress, and completed, according to the completion rate range. This generates a construction status classification result. Based on the construction status classification results, a time-series progress database is constructed. Time information extracted from progress logs is aggregated by day, week, and month, and the cumulative completed work volume for each time dimension is calculated. Based on the cumulative completed work volume, exponential smoothing is used to predict the construction progress data, resulting in a construction progress trend curve and construction cycle assessment results. When extracting key construction information from progress logs using word frequency statistics, if a log text contains a description such as "25 meters of excavation work on the pile foundation of a bridge on a certain road section," a construction-specific dictionary can identify keywords such as "pile foundation," "excavation," and "25 meters." "Pile foundation" matches the construction area type, "excavation" indicates the construction operation type, and "25 meters" represents the construction volume value. This creates a structured data record containing the area, operation, and value. In the geographic information processing of construction areas, the engineering geographic information database stores pre-measured pile foundation coordinate data, such as the coordinates for pile number K15+360 corresponding to 116.5° East longitude and 39.6° North latitude. This data is also associated with geological exploration data for these locations, such as rock formation type and bearing capacity. Through this correlation, a complete set of construction area characteristic data is formed. For earthwork excavation, for example, a circular pile foundation with a 10-meter excavation depth and a 3-meter radius excavation area is calculated to be 282 cubic meters. These quantitative construction indicators reflect the actual construction progress. Based on this data, the construction progress prediction model can learn construction patterns and predict the completion rate of this type of pile foundation. For bridge pile foundation projects, the construction status quantification rule classifies a completion rate of 0-30% as not started, 30%-80% as in progress, and 80%-100% as completed, enabling accurate assessment of construction progress. This classification method takes into account the characteristics of pile foundation construction, such as the initial drilling and hole cleaning accounting for 30% of the workload, and pouring concrete accounting for 50% of the workload.In processing time-series progress data, daily construction log records are compiled to determine the number of pile foundations completed and the amount of work completed each day. This data is then accumulated weekly to generate weekly progress data. For example, in the first week, three pile foundations totaling 847 cubic meters of earthwork excavation were completed, while in the second week, four pile foundations totaling 1,128 cubic meters of earthwork were completed. An exponential smoothing prediction model is developed based on this time-series data to predict the construction progress for the next cycle, thereby rationally allocating construction resources. For large bridge projects, pile foundation construction is often carried out simultaneously in different areas. Analysis of construction progress data for each area reveals a correlation between construction efficiency and geological conditions. For example, in soft soil areas, the construction period is generally extended by 20%. This data provides a basis for optimizing subsequent construction plans. Construction period assessment results show that under normal geological conditions, the construction period for a single pile foundation is five days, but under complex geological conditions, it can be extended to six days.

[0027] Step S102: Identify the semantic relationship between the excavation area, completed volume, and time triples based on the construction progress information set, and determine the spatiotemporal semantic representation.

[0028] Based on the pre-labeled construction log, excavation area description words, engineering quantity values and construction time information are obtained, and the conditional random field method is used to identify them to obtain labeled structured text data; for the structured text data, a word vector-based method is used to identify the area name, area value, and depth value to form an initial entity set; the initial entity set is standardized using semantic relationship annotation rules, and the standardization process obtains normalized entity data through regional affiliation rules, engineering quantity measurement rules and time format rules; vector embedding representation is performed based on the normalized entity data, and a relationship including triples of excavation area, completed volume and time is established to determine the spatiotemporal semantic representation.

[0029] For example, text features such as word frequency, part of speech, and location are extracted from a pre-annotated construction log sample library. Conditional random field methods are used to identify excavation area descriptors, project quantity values, and construction time information in the logs, generating labeled structured text data. Semantic features are extracted from the structured text data, and a word vector-based method is used to identify entity information such as excavation area names, area values, depth values, volume values, and construction dates, forming an initial entity set. Semantic relationship annotation rules are formulated for this initial entity set, including rules for regional affiliation, project quantity measurement, and time format. Entities are standardized based on these rules to generate normalized entity data. An entity-relationship database is constructed using a knowledge graph storage method, with nodes representing entity information and edges representing relationships between entities. A relational network consisting of triples of region, volume, and time is established. Deep learning embedding is performed on the entity nodes in the relational network to generate fixed-dimensional entity vector representations. The entity vectors contain regional, volume, and time feature dimensions. A semantic similarity matrix is calculated based on the entity vectors, with the matrix elements representing the degree of association between entities. The relationship strength between entity pairs is calculated using attention weights. Entity relationships are filtered based on their relationship strength values, and entity pairs whose relationship strength exceeds a preset threshold are selected to construct a spatiotemporal semantic graph, achieving a three-dimensional representation of associations across regions, volumes, and time. In construction progress logs, text feature extraction can involve multiple dimensions. Taking word frequency statistics as an example, the term "foundation pit" appears frequently in descriptions of excavation areas. Part-of-speech tagging indicates that it is a noun, and positional features indicate that it often appears at the beginning of sentences. These features collectively point to its potential as a regional descriptor. Furthermore, numerical features, such as the combination of numbers and units in "500 cubic meters," are identified as construction quantity information using a conditional random field approach. During semantic feature recognition, a construction log entry states, "Foundation pit No. 1: depth 8 meters, excavation area 625 square meters, earthwork excavation 5,000 cubic meters, construction date March 15, 2024." Using word embedding methods, we can identify the area name "Foundation pit No. 1," the depth value "8 meters," the area value "625 square meters," the volume value "5,000 cubic meters," and the time information "March 15, 2024." Semantic relationship annotation rules define how entities are associated. For example, regional affiliation rules specify the correspondence between foundation pit numbers and work area numbers; "Foundation Pit No. 1 is located in Work Area A" indicates the hierarchical relationship of regions. Quantity measurement rules define the conversion relationships between different units, facilitating the standardization of measurement standards. Time format rules convert "March 15th" to the standard format "2024-03-15." In knowledge graph construction, entity nodes store foundation pit information, such as "Foundation Pit No. 1," with attributes including characteristic values such as depth and area. Edges represent relationships between entities, such as "Excavation Volume" connecting region nodes with volume nodes, and "Construction Time" connecting region nodes with time nodes, forming a complete triplet network.The deep learning embedding process transforms entities in the graph into 300-dimensional vectors, with 100 dimensions representing regional features such as location and type, 100 dimensions representing volumetric features such as depth and area, and 100 dimensions representing temporal features such as year, month, and day. This vector representation captures the underlying characteristics of the entities. Semantic similarity calculations show that the cosine similarity of entity vectors in adjacent regions reaches 0.85, indicating strong correlation between spatially close regions. The temporal similarity of areas under construction during the same period is 0.92, reflecting the synchronization of construction processes. When the relationship strength exceeds a preset threshold of 0.8, the corresponding entity pair is retained in the spatiotemporal semantic graph. Furthermore, in a real-world application, three foundation pit areas at a construction site constitute a construction unit. The spatial distance between the pits is less than 50 meters, the construction time overlap reaches 80%, and the volume difference is within 20%. These characteristics manifest as high similarity between entities in vector space, forming strongly connected entity clusters in the semantic graph.

[0030] A regional topological structure is constructed based on the triples including excavation area, completed volume and time. The regional topological structure reflects the volume transfer relationship between adjacent excavation areas. The volume change gradient is calculated based on the regional topological structure. The graph neural network is used to fuse the regional topological structure and the volume change gradient to obtain a spatiotemporal semantic representation.

[0031] Based on the construction progress text sequence, the excavation area identifier, volume value and timestamp are extracted through the long short-term memory network to obtain the construction information triplet; for the regional latitude and longitude information in the construction information triplet, the inter-region distance matrix is calculated, and the adjacent region pairs are judged through the inter-region distance matrix to obtain the adjacent region boundary overlap ratio; based on the adjacent region pairs and the boundary overlap ratio, the regional volume change value is accumulated within a preset time window to obtain the unit distance volume change rate feature; the unit distance volume change rate feature is used to train the graph convolutional network, and the regional topological features are extracted through the graph convolutional network to obtain the regional node representation vector, and the regional transfer matrix is constructed to generate the spatiotemporal semantic representation.

[0032] For example, the progress information set is aligned according to the 24-hour time scale, and the text sequence is processed through the long short-term memory network to extract the excavation area name, the engineering quantity value in cubic meters, and the construction time point of year, month, day, hour and minute, forming a standard triple containing the area identifier, volume value, and timestamp. Based on the latitude and longitude information of the excavation area in the triple, the distance matrix between regions is calculated. Two regions with a distance less than a preset threshold are judged to be adjacent regions. The boundary overlap ratio is calculated by dividing the common length of the construction boundary lines of the adjacent regions by the total boundary length. For adjacent region pairs in the regional topology structure, a 7-day time window is set, and the regional volume change value is accumulated within the window. The volume change rate per unit distance is calculated to generate feature data including volume change rate, boundary overlap ratio, and regional distance. A three-layer graph convolutional network is used to extract regional topological features. The output dimensions of each layer of graph convolution are set to 128, 64, and 32 respectively. The features are reduced in dimensionality through the maximum pooling layer to generate regional node representation vectors. A transfer matrix is constructed based on the regional node representation vectors to record the volume transfer relationships between regions. The transfer matrix is then subjected to eigendecomposition, and the main eigenvectors are extracted as regional transfer features. The regional transfer features are concatenated with the volume change features, and the features are fused using a fully connected layer. The feature vectors are reorganized according to the three dimensions of region, volume, and time to generate a fixed-dimensional spatiotemporal semantic representation. Timestamp alignment is essential for processing construction progress information. Raw progress logs may contain ambiguous time descriptions such as "morning" and "afternoon." These records are standardized by converting them into specific time periods in a 24-hour clock.

[0033] For example, "Morning construction" can be converted to a timestamp of "08:00-12:00," allowing all progress data to be compared and analyzed on the same timescale. Excavation area identifiers are formatted as "A1-01," where "A1" represents the work area number and "01" represents the pit number. Work quantities are recorded in cubic meters, with timestamps using a format such as "2024-03-15 10:30." For analysis of regional spatial relationships, the locations of each construction area are determined using latitude and longitude coordinates. For example, the center of the A1-01 pit is located at 30.5 degrees north latitude and 114.3 degrees east longitude. The distance between adjacent pits is calculated based on these coordinates. Two pits are considered adjacent if the distance between them does not exceed 50 meters. Regarding the calculation of boundary overlap, considering the actual pit geometry, when two circular pits are involved, the ratio of the length of the overlapping boundary to the respective perimeters is used as the metric for overlap. Volume change features are extracted using a 7-day sliding window, recording changes in earthwork excavation volume over that period. For example, the excavation volume of the A1-01 foundation pit increased from 0 to 5,000 cubic meters between March 15 and March 21. The excavation volume of the A1-02 foundation pit, 50 meters away, increased from 0 to 4,800 cubic meters during the same period. The distance-normalized volume change rate reflects the spatial correlation of construction progress. In graph convolutional network processing, each construction area is represented as a node in the graph. Node features include attributes such as excavation depth and area, while edge features include information such as inter-area distance and overlap. The feature dimension of the three-layer graph convolution is gradually reduced, from the original 256-dimensional feature to 128-dimensional, 64-dimensional, and finally to a compact 32-dimensional representation. The regional transfer matrix captures the propagation of volume changes. For example, once excavation of the A1-01 foundation pit is completed, construction of the adjacent A1-02 foundation pit immediately begins, reflecting the spatial dependence of the construction sequence. The eigenvalue decomposition of the transfer matrix extracts the main patterns in the construction sequence, and the eigenvector corresponding to the maximum eigenvalue reflects the main transmission direction of volume changes. During the feature fusion stage, 32-dimensional regional node features are concatenated with 16-dimensional transfer features to form a 48-dimensional vector. This vector is then subjected to dimensionality reduction using a 32-node fully connected layer. The resulting 24-dimensional spatiotemporal semantic representation consists of 8 dimensions encoding regional attributes, 8 dimensions representing volume features, and 8 dimensions retaining temporal information. This feature representation preserves the core information about the temporal evolution of construction progress, its spatial distribution, and regional correlations.

[0034] Step S103: obtain the engineering design drawings and establish a BIM model containing a preset earthwork logical body. Combine the spatiotemporal semantic representation and the established BIM model, map the excavation area in the semantic relationship diagram with the earthwork logical body in the BIM model using a unique ID, associate the progress information with the model entity, and obtain the earthwork logical body.

[0035] According to the engineering drawing data, the three-dimensional earthwork units are divided into blocks according to the depth value, area value and elevation value to obtain the earthwork logical body with the area number-layer number-serial number identifier; the three-dimensional coordinate value of the center point and the boundary vertex coordinate value of the earthwork logical body are extracted, and the Euclidean distance value between the center points of any two earthwork logical bodies is calculated to determine that they are adjacent earthwork bodies when the Euclidean distance value is less than the preset distance threshold; a convolutional neural network is used to extract features of the spatiotemporal semantic representation to obtain a region representation vector consistent with the spatial feature dimension of the earthwork body, and the cosine similarity value between the region representation vector and the spatial feature of the earthwork body is calculated; the mapping relationship between the region and the earthwork body is screened according to the cosine similarity value to generate the earthwork logical body.

[0036] For example, three-dimensional earthwork units are created based on engineering drawing data. These units are then divided into blocks based on depth, area, and elevation, generating independent logical earthwork volumes. A unique identifier in the format of "area number-layer number-sequence number" is generated based on the construction area coding rules to identify each logical earthwork volume. Spatial feature parameters are extracted from the logical earthwork volumes, including the three-dimensional coordinates of the center point, a list of boundary vertex coordinates, excavation depth, and excavation area. The Euclidean distance between the center points of any two logical earthwork volumes is calculated; any distance less than a preset threshold is considered adjacent. An earthwork spatial network is constructed based on the relationships between adjacent earthwork volumes. The set of adjacent earthwork volume numbers for each earthwork volume is recorded, generating earthwork spatial relationship data containing node connectivity relationships. A convolutional neural network is used to extract regional feature vectors from the spatiotemporal semantic representation. These feature vectors are then transformed to generate region representation vectors that match the spatial feature dimensions of the earthwork volume. The cosine similarity between the region representation vectors and the spatial features of the earthwork volume is calculated to generate a similarity score matrix between the regions and the earthwork volume. The highest similarity value is used as the matching threshold. The mapping relationship between regions and earthwork volumes is filtered based on a matching threshold. Progress information from the semantic graph is written into the earthwork volume attribute table based on this mapping relationship. The attribute table contains fields for start time, completion time, and volume completed. Earthwork volumes with written progress information are sorted by start time and, combined with adjacency constraints, a construction sequence for logical earthwork volumes that satisfies the construction sequence is generated. During earthwork excavation at a construction site, the division of three-dimensional earthwork units requires consideration of multiple parameters. For example, in a deep foundation pit project, the basement excavation depth is 12 meters. Depth layers are divided every 3 meters, forming four vertical layers. Each layer is divided into areas based on the structural column grid spacing of 30 meters x 30 meters, generating independent logical earthwork volumes. Earthwork volumes are uniquely identified using the "A1-L2-01" format, where A1 represents the construction area, L2 represents the second layer, and 01 represents the first earthwork volume in that layer. Extracting the spatial features of a logical earthwork volume involves several key parameters, such as the center point coordinates (x=100, y=200, z=-6) representing the volume's position in the site coordinate system. The list of boundary vertex coordinates captures the volume's geometric shape. The excavation depth of 3 meters and the area of 900 square meters together define the volume's scale. Two volumes are considered adjacent when the center point distance is less than 45 meters. This threshold is determined based on the operating radius of the construction machinery. In the earthwork spatial network, volume A1-L2-01 is connected to four adjacent volumes: A1-L2-02 and A1-L2-04 on the same level, and A1-L1-01 and A1-L3-01 on the upper and lower levels. This network structure reflects the spatial constraints of earthwork excavation, requiring coordination of the excavation sequence between adjacent volumes. After dimensionality transformation, the regional feature vectors contain three types of features: location, scale, and depth. Similarity is calculated with the spatial features of the earthwork volume.When the calculated similarity score between a construction area and earthwork volume A1-L2-01 is 0.92, exceeding the preset threshold of 0.85, a correspondence is confirmed between the two. After the progress information is written to the earthwork volume attribute table, each earthwork volume contains specific construction time information. For example, A1-L2-01 has a start date of March 15, 2024, a completion date of March 20, 2024, and a completed volume of 2,700 cubic meters. Earthwork volumes are sorted based on these time attributes, taking into account construction constraints between adjacent earthwork volumes. For example, construction of A1-L2-02 can only begin after A1-L2-01 is completed. The generation of the earthwork construction sequence also considers vertical layering constraints. Excavation of the upper earthwork layer must begin after the lower layer is completed. For example, construction of the A1-L2 layer can only begin after the excavation of the four earthwork volumes in the A1-L1 layer is complete. In this way, an earthwork excavation sequence is generated that satisfies both spatial adjacency and construction process requirements. This sequential arrangement ensures construction safety and efficiency during foundation pit excavation.

[0037] Extract the three-dimensional coordinate range attributes from the set of earthwork logical bodies, perform spatial overlay calculation with the spatiotemporal codes in the semantic relationship graph, establish a mapping relationship pair between the spatiotemporal codes and the logical body ID based on the overlay calculation results, verify the spatial consistency of the mapping relationship pair, align the timestamps with the progress plan data table after the verification, and update the progress status attributes in the set of earthwork logical bodies.

[0038] Extract the vertex coordinate array from the earthwork logical body, and use the octree spatial index method to spatially divide the vertex coordinate array to obtain bounding box data containing the minimum point coordinates and the maximum point coordinates; extract the region boundary point sequence based on the bounding box data, construct a three-dimensional grid unit, and perform an intersection operation on the three-dimensional grid unit to obtain spatial overlapping volume data; calculate the ratio of the overlapping volume to the total volume of the earthwork body for the spatial overlapping volume data, and if the ratio is greater than a preset threshold, record the correspondence between the earthwork logical body identifier and the spatiotemporal coding to obtain mapping relationship data; obtain the construction process definition based on the mapping relationship data, and write the planned start time, planned completion time, and current construction status into the attribute field of the earthwork logical body by comparing with the spatiotemporal coding mapping table.

[0039] Exemplarily, an array of vertex coordinates is extracted from the earthwork logical volume. An octree spatial index is constructed based on the vertex coordinates, and the minimum and maximum coordinates of the bounding box are recorded to generate spatial extent description data for the earthwork logical volume. A sequence of regional boundary points is extracted from the spatiotemporal code in the semantic relationship graph. A three-dimensional mesh is constructed using the Delaunay triangulation method. The spatial overlap volume is calculated through grid cell intersection operations to generate an overlapping region dataset. Based on the overlapping region dataset, the ratio of the overlapping volume to the total volume of the earthwork volume is calculated. If the ratio exceeds a preset threshold, the correspondence between the earthwork logical volume identifier and the spatiotemporal code is recorded to generate a preliminary mapping result. A random forest method is used to classify the preliminary mapping results. Input features include overlap ratio, spatial distance, and elevation difference, and the output is a confidence score for the mapping relationship. Mapping relationships are screened based on the confidence score, and high-confidence mapping pairs are retained according to a preset confidence threshold to establish a final spatiotemporal code mapping table. Construction process definitions, including process number, process name, and planned duration, are obtained from the schedule data table to generate a standardized schedule data structure. By comparing the time-space coding mapping table, the schedule data is written into the corresponding earthwork logical volume attribute fields. These attribute fields contain the planned start time, planned completion time, and current construction status. A process dependency graph is established based on the construction process definition. The set of earthwork logical volumes is sorted according to the process sequence constraints, and the construction sequence numbers of the logical volumes are updated. In earthwork projects, the spatial positioning of earthwork logical volumes requires complex three-dimensional coordinate representation. For example, in a basement foundation pit, the eight vertices of an earthwork volume are located at different elevations, forming an irregular hexahedron. Using an octree spatial index, the space is divided into a progressive cubic grid with a minimum grid edge length of 1 meter, enabling rapid retrieval of the spatial location of the earthwork volume. In the area boundary representation, the boundary of construction area A1 is composed of 20 spatial points forming a closed polygon, each containing a 3D coordinate value, such as point P1 (100, 200, -6). Delaunay triangulation decomposes this polygon into 18 triangular units, forming a unified 3D grid representation. When two areas are spatially superimposed, the overlapping volume is obtained by intersecting the grid units. The calculation of spatial overlap reflects the correspondence between the earthwork logical volume and the construction area. For example, the overlap volume between earthwork T1 and construction area A1 is 450 cubic meters, accounting for 90% of T1's total volume of 500 cubic meters. This exceeds the preset threshold of 80%, indicating a significant spatial correspondence between T1 and A1. This quantitative spatial relationship measurement provides a basis for subsequent mapping verification. In addition to the overlap ratio, spatial distance and elevation difference are also considered in the classification of mapping relationships. For example, the center point distance between earthwork T1 and construction area A1 is 5 meters, and the elevation difference is 0.5 meters. These features together determine the reliability of the mapping relationship. The random forest classifier learns feature patterns based on historical annotated data and scores new mapping pairs. Mapping relationships with a confidence level exceeding 0.9 are retained. The timing constraints of the construction process are reflected in the schedule data.Taking foundation pit excavation as an example, the process definition includes setting out and positioning (process number W001), earthwork excavation (W002), and base preparation (W003). Each process has a clear duration requirement. The excavation process is scheduled to last for 5 days, and this time constraint is written into the attribute fields of the corresponding earthwork. The process dependencies form a directed acyclic graph: setting out and positioning must precede earthwork excavation, which in turn precedes base preparation. This dependency determines the construction sequence of earthworks. For example, earthwork T1 has a construction sequence number of 1, while its adjacent earthwork T2 has a construction sequence number of 2, ensuring seamless process connection during construction. When the construction status of an earthwork changes, such as T1 changing from "Not Started" to "Under Construction," the system checks the status of its adjacent earthworks. If T2 is still in the "Not Started" state and its predecessor process T1 has not yet completed, T2 will remain in the "Not Started" state, reflecting the constraints imposed by process dependencies on the construction sequence.

[0040] In step S104, the geometric information of each target earthwork logical body is extracted from the set of earthwork logical bodies, and combined with the corresponding elevation data extracted from the BIM model, the design engineering quantity of each associated logical body within the corresponding preset elevation range is calculated in real time to generate a design quantity data set.

[0041] A triangular mesh surface is constructed according to the vertex coordinates of the earthwork logical volume, and the triangular mesh surface is voxelized to obtain voxel mesh data; the earthwork excavation elevation control lines are extracted from the design drawings, and the elevation control lines are interpolated to obtain the elevation control surface dataset; the voxel mesh data is layered and segmented according to the elevation control surface dataset, and the segmented voxel units are spatially grouped using the nearest neighbor clustering method; for the spatially grouped voxel units, the loose coefficient is obtained according to the soil classification table, and the convex hull algorithm is used to reconstruct the earthwork boundary surface of each layer to generate the design quantity dataset.

[0042] For example, a triangular mesh surface is constructed based on the vertex coordinates of the earthwork logical volume. The mesh side length is set to the smallest unit of terrain undulation. The normal vector and area parameters of each triangular facet are extracted, and the irregular earthwork volume is voxelized using the octree partitioning method to generate voxel mesh data with the side length as the basic unit. Earthwork excavation elevation control lines are extracted from the design drawings. A continuous elevation control surface is generated using the Kriging interpolation algorithm. The spatial position and elevation value of each control point on the surface are recorded to establish an elevation constraint dataset. Based on the elevation constraint dataset, the earthwork voxel grid is layered and segmented. Elevation intervals are divided according to the design elevation values, and the number and spatial distribution of voxel units within each interval are recorded. For each voxel unit in each elevation interval, a feature vector containing spatial coordinates, elevation values, and neighbor relationships is extracted. The voxels are spatially grouped using the nearest neighbor clustering method. The boundaries of the grouped voxel units are identified, and the convex hull algorithm is used to reconstruct the earthwork boundary surface of each layer, and the earthwork volume value enclosed by the surface is calculated. Based on geological survey data, a soil classification table was established, including soil parameters for clay, sand, and gravel soils. A database was also established that correlated soil types with bulk coefficients. The design volume of each layer of earthwork was calculated using the soil classification table. The design volume of earthwork was calculated by multiplying the volume of the earthwork by the bulk coefficient of the corresponding soil type, generating a statistical table of layered earthwork quantities. The layered earthwork quantities were summarized to create an earthwork data record containing the earthwork number, excavation elevation, soil type, and design volume, generating a data set for the designed earthwork quantity. Surface feature extraction of irregular earthwork volumes is crucial in geometric modeling of earthwork projects. For example, a deep foundation pit project has an irregular polygonal surface. Triangulated meshing is used to divide the surface into triangular face elements with a side length of 2 meters. Each face element has a normal vector and area attribute. An octree method was used to partition the earthwork volume into 1-cubic-meter voxel elements, achieving the conversion from surface features to volume features. Elevation control is a key constraint in earthwork excavation. In this foundation pit project, the excavation elevation ranged from -12 meters to 0 meters. A continuous elevation surface was generated between known control points using the Kriging interpolation algorithm. The control points were arranged according to the terrain's dynamics, with denser points in areas of high relief and more sparse points in flatter areas. This resulted in an elevation constraint network consisting of 2,000 control points. A layered design is essential for accurately calculating earthwork volumes. The excavation depth was divided into 3-meter intervals, creating four elevation layers: -12 meters to -9 meters, -9 meters to -6 meters, -6 meters to -3 meters, and -3 meters to 0 meters. Each layer contains voxel units with clear spatial location information. The connectivity between adjacent voxels forms a spatial topological network, facilitating subsequent grouping calculations. For earthwork boundary processing, the nearest neighbor clustering method grouped voxels with similar spatial locations. The earthwork boundary for each elevation layer was reconstructed using the convex hull algorithm, resulting in a closed boundary surface. The first layer, from -12m to -9m, has a volume of 5,000 cubic meters and a boundary area of 1,800 square meters. Soil classification directly affects the engineering quantity calculation.According to geological survey data, the foundation pit has three main soil types: clay in the surface layer with a looseness coefficient of 1.3; sand in the middle layer with a looseness coefficient of 1.25; and pebble soil in the bottom layer with a looseness coefficient of 1.2. The distribution patterns of different soil types clearly correspond to elevation levels. In project quantity calculations, the designed volume is calculated by multiplying the volume of each layer by its corresponding looseness coefficient. For example, the layer from -12 meters to -9 meters consists of pebble soil with a volume of 5,000 cubic meters. After accounting for the looseness coefficient of 1.2, the designed volume is 6,000 cubic meters. This layered calculation method accounts for the impact of soil quality variations on project quantities. Earthwork data records contain complete project quantity information. For example, earthwork unit numbered TF001, excavated from elevations -12 meters to -9 meters, has a pebble soil type and a designed volume of 6,000 cubic meters. This structured data recording method provides a foundation for subsequent construction progress management. Through this refined earthwork quantity calculation system, accurate conversion from geometric features to engineering quantities is achieved.

[0043] Step S105: Evaluate the geological complexity level based on the distribution characteristics of the geological complexity adjacent to the excavation area, and dynamically adjust the allowable deviation based on the complexity level to calculate a dynamic deviation threshold set. When the difference between the accumulated or single reported actual completed volume data and the design volume data set exceeds the threshold of the corresponding area and complexity in the dynamic deviation threshold set, the warning condition is triggered, and the corresponding earthwork logical body is marked in the BIM model to obtain a warning result.

[0044] The top and bottom elevations of the rock and soil layers, the rock and soil type code, the standard penetration blows and the water content parameters in the borehole sampling data are obtained, and the support vector machine method is used to obtain the rock and soil layer classification results; based on the rock and soil layer classification results and the borehole sampling data, the geological parameters between the borehole positions are processed by the Kriging spatial interpolation method to obtain the geological distribution surface covering the construction area; the geological distribution surface is discretized by the square grid division method, and the mean and variance data of the grid unit geological parameters are recorded according to the grid side length; the coefficient of variation is calculated for the variance data of the grid unit geological parameters. If the coefficient of variation is less than a preset first threshold, it is divided into a low complexity area; if the coefficient of variation is greater than the first threshold and less than a preset second threshold, it is divided into a medium complexity area; if the coefficient of variation is greater than the second threshold, it is divided into a high complexity area.

[0045] For example, based on the borehole sampling data, geotechnical stratification information is extracted, including the top and bottom elevations of the geotechnical layer, geotechnical type code, standard penetration number, and water content parameters. The support vector machine method is used to classify the geotechnical layer into hard soil layer, soft soil layer, and sand and gravel layer, and a geological distribution feature database is established. The geological parameters between the borehole locations are processed continuously using the Kriging spatial interpolation method. The interpolation parameters include geotechnical type, standard penetration number, and water content, and a geological distribution surface covering the entire construction area is generated. The geological distribution surface is discretized using the square grid division method. The grid side length is set to one-quarter of the borehole spacing, and the mean and variance of the geological parameters of each grid unit are recorded. The coefficient of variation is calculated based on the variance of the geological parameters of the grid unit. A coefficient of variation less than 0.2 is classified as low complexity, 0.2 to 0.5 as medium complexity, and greater than 0.5 as high complexity. The benchmark deviation value is extracted from the industry specification. The low-complexity area takes 1.0 times the benchmark value, the medium-complexity area takes 1.2 times, and the high-complexity area takes 1.5 times. A dynamic deviation threshold table for each region is generated. The difference between the actual completed volume and the designed volume is divided by the designed volume to obtain the deviation rate, which is judged against the dynamic deviation threshold table. If the deviation rate exceeds the threshold, the exception number, occurrence time, and deviation value are written to the earthwork logical body attribute table, and the warning level field is updated. The earthwork logical body is color-coded according to the warning level. The low-level warning mark is yellow, the medium-level warning mark is orange, and the high-level warning mark is red. Warning display data is generated. In geotechnical engineering surveys, drilling sampling data reflects the basic characteristics of the stratum. Taking a deep foundation pit project as an example, the drilling depth reaches 20 meters. The geotechnical parameters recorded at different depths include key indicators such as standard penetration count and water content. The upper layer (0-5 meters) is a hard soil layer with a standard penetration number of 25-30 blows and a moisture content of 15%; the middle layer (5-12 meters) is a soft soil layer with a standard penetration number of 4-8 blows and a moisture content of 35%; the lower layer (12-20 meters) is a sand and gravel layer with a standard penetration number greater than 50 blows. During the spatial interpolation process, 25 boreholes were laid out on the project site, with a spacing of 40 meters. A continuous geological distribution surface was generated using the Kriging interpolation method, with the interpolation grid size set to 10 meters, which not only ensured data continuity but also captured the detailed characteristics of geological changes. The resulting geological distribution surface clearly demonstrates the spatial distribution patterns of different geotechnical layers. Gridding is the basis for assessing geological complexity. Each 10-meter square grid cell records the statistical characteristics of geotechnical parameters, and the coefficient of variation is calculated by calculating the parameter variance. In the transition zone between the soft soil layer and the sand and gravel layer, due to the sudden change in lithology, the coefficient of variation reaches 0.6, indicating a high complexity zone. The coefficient of variation in the pure soft soil layer is 0.3, indicating a medium complexity zone. The coefficient of variation in the single sand and gravel layer is only 0.15, indicating a low complexity zone. Deviation control based on geological complexity fully considers construction risks. The baseline deviation value stipulated by industry standards is 5% of the design value, and is dynamically adjusted for different complexity zones.For example, for an area with a designed excavation volume of 1,000 cubic meters, the tolerance is 50 cubic meters for low-complexity areas, 60 cubic meters for medium-complexity areas, and 75 cubic meters for high-complexity areas. During actual construction monitoring, the designed excavation volume for a high-complexity area was 2,000 cubic meters, but the actual completed volume was 2,200 cubic meters, representing a 10% deviation rate, exceeding the 7.5% tolerance threshold for that area. The earthwork logical volume in that area was immediately marked red, with key information such as the anomaly number AC001, the occurrence date 2024-03-15, and the deviation value of 200 cubic meters. The warning display utilizes a multi-level color coding system. Within the entire construction site earthwork model, areas with a deviation rate of 5%-7.5% display yellow, indicating a low-level warning; areas with a deviation rate of 7.5%-10% display orange, indicating a medium-level warning; and areas with a deviation rate exceeding 10% display red, indicating a high-level warning. This tiered warning mechanism enables construction managers to quickly identify problem areas.

[0046] Step S106: Based on the early warning result, the state attributes of the earthwork logical body in the BIM model are updated, and the timestamp and difference of the deviation are combined to generate a dynamically updated BIM model state, extract the spatial distribution of the earthwork logical body in the BIM model, and generate a construction path.

[0047] Obtain the warning level mark and deviation value recorded in the earthwork logical body attribute table, where the deviation value represents the difference between the actual volume and the designed volume; generate a deviation value time series based on the deviation value, and use a long short-term memory network to calculate the autocorrelation coefficient and the cross-correlation coefficient of the deviation value time series to obtain multi-dimensional deviation statistical data; run an exponential smoothing algorithm on the multi-dimensional deviation statistical data to calculate the deviation growth rate and acceleration parameters to obtain a deviation trend prediction curve; calculate the risk score based on the deviation trend prediction curve and the process dependency table, and use a dynamic programming algorithm to optimize and sort the risk scores to obtain a construction path that meets the process constraints.

[0048] For example, a warning level tag (low, medium, and high) is written into the earthwork logical volume attribute table. The deviation direction type includes overbreak and underbreak, and the deviation value records the difference between the actual volume and the designed volume. A state change record table is created, containing the attribute identifier, earthwork number, and update timestamp. The state change record table is sorted by timestamp, and the warning level, deviation type, and deviation value at each time point are extracted. The autocorrelation coefficient and cross-correlation coefficient of the deviation value sequence are calculated using a long short-term memory network. The cumulative effect of the deviation at different time scales is calculated based on the correlation coefficient, and a multi-dimensional deviation statistics table is created, including daily, weekly, and monthly cumulative values. An exponential smoothing algorithm is used to perform trend analysis on the multi-dimensional deviation data, calculating the deviation growth rate and acceleration, and generating a deviation trend prediction curve. The construction process number is extracted from the earthwork volume attribute record. Based on the process flow chart, a process dependency table is created, recording the preceding process, subsequent process, and the time constraint between processes. Combining the process dependency and the deviation trend prediction curve, a risk score is calculated for each earthwork volume. The risk score is determined by the degree of deviation, frequency of occurrence, and trend prediction. Construction processes are prioritized based on risk scores, and a dynamic programming algorithm is used to calculate the optimal construction path that satisfies process constraints. The optimization objective is to minimize the continuous construction duration in high-risk areas. The optimized construction path is converted into a construction sequence table, containing earthwork number, planned start time, and planned completion time. A construction schedule is generated based on the construction sequence. Dynamic updates to warning status in earthwork construction involve multi-dimensional data analysis. For example, the status change record for the earthwork logical volume TF001, during the excavation of an underground garage foundation pit, shows that an overexcavation warning was issued on March 15th, with a deviation of 100 cubic meters, triggering a low-level warning. By March 18th, the deviation increased to 200 cubic meters, raising the warning level to medium. On March 20th, the deviation reached 300 cubic meters, further escalating to a high-level warning. Deviation sequence analysis reflects the evolution of problems during construction. Long-short-term memory network calculations revealed that the deviation value in the TF001 area increased by 100 cubic meters every three days, showing a clear linear growth trend. Adjacent areas TF002 and TF003 also exhibited a similar deviation growth pattern, with an autocorrelation coefficient reaching 0.85, indicating that the deviation problem exhibits spatial propagation. Analysis of the cumulative effect of deviations revealed that daily cumulative data showed faster deviation growth on weekdays and slower growth on weekends. Weekly cumulative data revealed a 300 cubic meter deviation in the first week and a 500 cubic meter deviation in the second. Monthly cumulative data forecasts indicated that if no action was taken, the deviation would exceed 1,000 cubic meters by the end of the month. A trend curve fitted using an exponential smoothing algorithm predicted a deviation growth rate of 35 cubic meters per day, with an accelerating upward trend. Process dependencies play a key role in construction optimization.Taking staged excavation as an example, the process flow chart shows that support pile construction (W001) must precede excavation (W002), with an interval of at least 7 days. After excavation is completed, base treatment (W003) must be carried out, with an interval of at least 3 days. Deviations in any process can affect the timing of subsequent processes. The risk score calculation takes into account multiple factors. Due to continuous overexcavation, the TF001 area has a deviation score of 0.8, an occurrence frequency score of 0.9, and a trend prediction score of 0.7, resulting in an overall risk score of 0.8, placing it in a high-risk area. To prevent risk contagion, the optimization solution generated by the dynamic programming algorithm appropriately postpones the construction of TF001 and its adjacent areas, inserting construction processes into low-risk areas. The final construction schedule reflects the risk management strategy. The original plan for the sequential construction of TF001, TF002, and TF003 was broken up. Construction of TF001 was then followed by construction of the low-risk area TF008. Construction of TF002 would proceed only after the deviation in TF001 was controlled. This staggered construction method reduces the continuous construction time in high-risk areas from 15 days to 5 days.

[0049] The present invention provides a BIM technology and engineering quantity intelligent verification system, which mainly includes:

[0050] The progress information processing module is used to perform word segmentation and part-of-speech tagging on the progress log, extract the excavation area, completed volume and timestamp of the day, and obtain the construction progress information set;

[0051] The semantic relationship recognition module is used to identify the semantic relationship between the excavation area, completed volume and time triples based on the construction progress information set, and determine the spatiotemporal semantic representation;

[0052] The model mapping module is used to obtain engineering design drawings and establish a BIM model containing a preset earthwork logical body. By combining the spatiotemporal semantic representation with the established BIM model, the excavation area in the semantic relationship diagram is mapped to the earthwork logical body in the BIM model using a unique ID, and the progress information is associated with the model entity to obtain the earthwork logical body.

[0053] The engineering quantity calculation module is used to extract the geometric information of each target earthwork logical body from the set of earthwork logical bodies, and combine it with the corresponding elevation data extracted from the BIM model to calculate the design engineering quantity of each associated logical body within the corresponding preset elevation range in real time to generate a design quantity data set;

[0054] The early warning analysis module is used to evaluate the geological complexity level based on the distribution characteristics of the geological complexity of the excavation area, and dynamically adjust the allowable deviation according to the complexity level. The dynamic deviation threshold set is calculated. When the difference between the accumulated or single reported actual completed volume data and the design volume data set exceeds the threshold of the corresponding area and complexity in the dynamic deviation threshold set, the early warning condition is triggered, the corresponding earthwork logical body is marked in the BIM model, and the early warning result is obtained;

[0055] The status update module is used to update the status attributes of the earthwork logical body in the BIM model according to the early warning results, generate a dynamically updated BIM model status based on the deviation occurrence timestamp and difference, extract the spatial distribution of the earthwork logical body in the BIM model, and generate a construction path.

[0056] Although the embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention.

Claims

1. A BIM technology and engineering quantity intelligent verification method, characterized in that: The method comprises: Perform word segmentation and part-of-speech tagging on the progress log to extract the excavation area, completed volume, and timestamp for the day to obtain a collection of construction progress information; Based on the construction progress information set, the semantic relationship between the excavation area, completed volume and time triples is identified to determine the spatiotemporal semantic representation; Obtain the engineering design drawings and establish a BIM model containing a preset earthwork logical body. Combine the spatiotemporal semantic representation with the established BIM model, map the excavation area in the semantic relationship diagram with the earthwork logical body in the BIM model using a unique ID, associate the progress information with the model entity, and obtain the earthwork logical body. Extract the geometric information of each target earthwork logical body from the set of earthwork logical bodies, calculate the design engineering quantity of each associated logical body within the corresponding preset elevation range, and generate a design quantity data set; The allowable deviation is dynamically adjusted according to the geological complexity level. When the difference between the actual volume data reported cumulatively or once and the design volume data set exceeds the allowable deviation, the warning condition is triggered, the corresponding earthwork logical body is marked, and the warning result is obtained; According to the early warning results, after updating the status attributes of the earthwork logical body in the BIM model, the spatial distribution of the earthwork logical body in the BIM model is extracted to generate the construction path.

2. The method according to claim 1, characterized in that The progress log is segmented and tagged with parts of speech, and the excavation area, completed volume and timestamp of the day are extracted to obtain a set of construction progress information, including: Extract keywords from the progress log text using a word frequency statistics method to establish a construction professional dictionary, which includes fields such as construction area name, excavation quantity value, and construction time; According to the construction area information in the construction professional dictionary, the latitude and longitude coordinate values and geological parameters of the construction area are obtained from a preset engineering geographic information database to obtain a construction area feature data set; Calculate the excavation depth and excavation area of the construction area based on the construction area characteristic data set to obtain construction quantitative index data; The random forest algorithm is used to analyze the construction quantitative indicator data, the construction completion rate is calculated by the least squares method, and the construction status classification results are divided according to the completion rate interval to form a construction progress information set.

3. The method according to claim 1, characterized in that The identifying of the semantic relationship between the excavation area, the completed volume, and the time triple based on the construction progress information set and determining the spatiotemporal semantic representation includes: Based on the pre-annotated construction logs, we obtain excavation area description words, engineering quantity values, and construction time information, and use the conditional random field method to identify them to obtain labeled structured text data. Using a word vector-based method to identify the region name, area value, and depth value for the structured text data to form an initial entity set; The initial entity set is standardized using semantic relationship annotation rules, wherein the standardization process obtains normalized entity data through regional affiliation rules, engineering quantity measurement rules, and time format rules; A vector embedding representation is performed based on the normalized entity data, a relationship including a triple of excavation area, completed volume and time is established, and a spatiotemporal semantic representation is determined.

4. The method according to claim 3, characterized in that Also includes: A regional topology is constructed based on the triples of excavation area, completed volume, and time. The regional topology reflects the volume transfer relationship between adjacent excavation areas. The volume change gradient is calculated based on the regional topology. A graph neural network is used to fuse the regional topology and volume change gradient to obtain a spatiotemporal semantic representation, which specifically includes: According to the construction progress text sequence, the excavation area identifier, volume value and timestamp are extracted through the long short-term memory network to obtain the construction information triple. Calculating an inter-region distance matrix for the regional longitude and latitude information in the construction information triplet, determining adjacent region pairs through the inter-region distance matrix, and obtaining an adjacent region boundary overlap ratio; According to the adjacent area pairs and the boundary overlap ratio, the regional volume change values are accumulated within a preset time window to obtain the unit distance volume change rate feature; The unit distance squared volume change rate feature is used to train a graph convolutional network, and the regional topological features are extracted through the graph convolutional network to obtain the regional node representation vector, and the regional transfer matrix is constructed to generate spatiotemporal semantic representation.

5. The method according to claim 1, characterized in that The engineering design drawings are obtained and a BIM model including a preset earthwork logical body is established. The excavation area in the semantic relationship diagram is mapped to the earthwork logical body in the BIM model by a unique ID, and the progress information is associated with the model entity to obtain the earthwork logical body, including: According to the engineering drawing data, the three-dimensional earthwork unit is divided into blocks according to the depth value, area value and elevation value to obtain the earthwork logical body with the area number-layer number-sequence number identifier; Extracting the three-dimensional coordinates of the center points and the coordinates of the boundary vertices of the earthwork logical bodies, and calculating the Euclidean distance between the center points of any two earthwork logical bodies to determine that the earthwork bodies are adjacent when the Euclidean distance is less than a preset distance threshold; A convolutional neural network is used to extract features from the spatiotemporal semantic representation, obtain a region representation vector that is consistent with the spatial feature dimension of the earthwork, and calculate the cosine similarity value between the region representation vector and the spatial feature of the earthwork; The mapping relationship between the region and the earthwork volume is screened according to the cosine similarity value to generate an earthwork logical volume.

6. The method according to claim 5, characterized in that Also includes: Extract the three-dimensional coordinate range attribute from the earthwork logical body set, perform spatial overlay calculation with the spatiotemporal code in the semantic relationship graph, establish a mapping relationship pair between the spatiotemporal code and the logical body ID based on the overlay calculation result, verify the spatial consistency of the mapping relationship pair, align the timestamp with the progress plan data table after the verification passes, and update the progress status attribute in the earthwork logical body set. Specifically, it includes: Extracting a vertex coordinate array from the earthwork logical volume, and performing spatial partitioning on the vertex coordinate array using an octree spatial indexing method to obtain bounding box data containing minimum point coordinates and maximum point coordinates; Extracting a region boundary point sequence according to the bounding box data, constructing a three-dimensional grid unit, and performing an intersection operation on the three-dimensional grid unit to obtain a spatial overlapping volume; Calculating the ratio of the overlapping volume to the total volume of the earthwork body for the spatial overlapping volume, and if the ratio is greater than a preset threshold, recording the corresponding relationship between the earthwork logical body identifier and the spatiotemporal code to obtain mapping relationship data; The construction process definition is obtained according to the mapping relationship data, and the planned start time, planned completion time, and current construction status are written into the attribute fields of the earthwork logical body by comparing with the spatiotemporal coding mapping table.

7. The method according to claim 1, characterized in that The method of extracting geometric information of each target earthwork logical volume from the set of earthwork logical volumes, calculating the design engineering quantity of each associated logical volume within a corresponding preset elevation range, and generating a design quantity data set includes: Constructing a triangular mesh surface according to the vertex coordinates of the earthwork logical body, and voxelizing the triangular mesh surface to obtain voxel mesh data; Extracting earthwork excavation elevation control lines from the design drawings, performing interpolation operations on the elevation control lines, and obtaining elevation control surface data sets; For the voxel grid data, hierarchical segmentation is performed according to the elevation control surface data set, and the segmented voxel units are spatially grouped using a nearest neighbor clustering method; For the spatially grouped voxel units, the loose coefficient is obtained according to the soil classification table, and the convex hull algorithm is used to reconstruct the boundary surface of each layer of earthwork to generate the design quantity data set.

8. The method according to claim 1, characterized in that The allowable deviation is dynamically adjusted according to the geological complexity level. When the difference between the accumulated or single reported actual volume data and the designed volume data set exceeds the allowable deviation, the warning condition is triggered, the corresponding earthwork logical body is marked, and the warning result is obtained, including: Obtain the top and bottom elevations, rock and soil type codes, standard penetration blows, and moisture content parameters of the rock and soil layers from the borehole sampling data, and use the support vector machine method to obtain the rock and soil layer classification results; Based on the rock and soil layer classification results and the borehole sampling data, the geological parameters between the borehole locations are processed by the Kriging spatial interpolation method to obtain a geological distribution surface covering the construction area; The geological distribution surface is discretized using a square grid division method, and the mean and variance data of the geological parameters of the grid units are recorded according to the grid side length; The coefficient of variation is calculated for the variance data of the geological parameters of the grid units. If the coefficient of variation is less than a preset first threshold, it is divided into a low-complexity area. If the coefficient of variation is greater than the first threshold and less than a preset second threshold, it is divided into a medium-complexity area. If the coefficient of variation is greater than the second threshold, it is divided into a high-complexity area.

9. The method according to claim 1, characterized in that After updating the state attributes of the earthwork logical body in the BIM model based on the early warning results, extracting the spatial distribution of the earthwork logical body in the BIM model and generating a construction path include: Obtaining the warning level mark and deviation value recorded in the earthwork logical body attribute table, wherein the deviation value represents the difference between the actual volume and the designed volume; Generate a deviation value time series according to the deviation value, calculate the autocorrelation coefficient and the cross-correlation coefficient of the deviation value time series using a long short-term memory network, and obtain multi-dimensional deviation statistical data; Running an exponential smoothing algorithm on the multi-dimensional deviation statistical data, calculating the deviation growth rate and acceleration parameters, and obtaining a deviation trend prediction curve; The risk scores are calculated based on the deviation trend prediction curve and the process dependency table, and the risk scores are optimized and sorted using a dynamic programming algorithm to obtain a construction path that meets the process constraints.

10. A BIM technology and engineering quantity intelligent verification system, characterized by: The system comprises: The progress information processing module is used to perform word segmentation and part-of-speech tagging on the progress log, extract the excavation area, completed volume and timestamp of the day, and obtain the construction progress information set; The semantic relationship recognition module is used to identify the semantic relationship between the excavation area, completed volume and time triples based on the construction progress information set, and determine the spatiotemporal semantic representation; The model mapping module is used to obtain engineering design drawings and establish a BIM model containing a preset earthwork logical body. By combining the spatiotemporal semantic representation with the established BIM model, the excavation area in the semantic relationship diagram is mapped to the earthwork logical body in the BIM model using a unique ID, and the progress information is associated with the model entity to obtain the earthwork logical body. The engineering quantity calculation module is used to extract the geometric information of each target earthwork logical body from the set of earthwork logical bodies, and combine it with the corresponding elevation data extracted from the BIM model to calculate the design engineering quantity of each associated logical body within the corresponding preset elevation range in real time to generate a design quantity data set; The early warning analysis module is used to evaluate the geological complexity level based on the distribution characteristics of the geological complexity of the excavation area, and dynamically adjust the allowable deviation according to the complexity level. The dynamic deviation threshold set is calculated. When the difference between the accumulated or single reported actual completed volume data and the design volume data set exceeds the threshold of the corresponding area and complexity in the dynamic deviation threshold set, the early warning condition is triggered, the corresponding earthwork logical body is marked in the BIM model, and the early warning result is obtained; The status update module is used to update the status attributes of the earthwork logical body in the BIM model according to the early warning results, generate a dynamically updated BIM model status based on the deviation occurrence timestamp and difference, extract the spatial distribution of the earthwork logical body in the BIM model, and generate a construction path.

Images