BIM-based project cost management method and system

Abstract

This application provides a BIM-based method and system for engineering cost management, relating to the field of engineering cost management technology. The method includes: collecting 3D point cloud data of components at the construction site, combining it with geometric attribute data and cost data from the BIM model to generate engineering cost data and establish cost geometric correlation information; performing geometric analysis on the point cloud data to extract 3D deformation data, and constructing a component relationship diagram based on the BIM model; mapping the deformation data and cost information to the relationship diagram, generating a cost deviation probability distribution through a dual-channel convolutional neural network, and finally using a reinforcement learning algorithm to generate a cost compensation strategy and automatically update the BIM database, thereby achieving dynamic adjustment of engineering costs. This application improves the adaptive control capability of engineering costs under construction geometric deformation scenarios.

Description

Technical Field

[0001] This application relates to the field of engineering cost management technology, and in particular to a BIM-based engineering cost management method and system. Background Technology

[0002] Construction cost management methods hold a crucial position in the construction industry, and their development directly impacts project cost control efficiency. With the widespread adoption of Building Information Modeling (BIM), digitally-based cost management is showing broad application prospects.

[0003] In existing technologies, construction cost management based on Building Information Modeling (BIM) mainly involves extracting component information from the model for cost calculation. Some methods identify discrepancies between the design model and on-site measurement data and adjust the cost budget based on preset rules.

[0004] These methods have limitations in matching cost adjustments with the actual deformation state when dealing with cost fluctuations caused by geometric deformation during construction. The accuracy of cost prediction in dynamic construction environments needs to be improved. Therefore, existing technologies suffer from insufficient matching between cost control and actual site conditions. Summary of the Invention

[0005] This application provides a BIM-based engineering cost management method and system to solve the problem of low adaptive adjustment capability of engineering cost in construction geometric deformation scenarios in the prior art.

[0006] To address the aforementioned technical problems, in a first aspect, this application provides a BIM-based engineering cost management method, comprising:

[0007] The system collects 3D point cloud data of components from the construction site, obtains geometric attribute data of components from the BIM model, and obtains design cost data and cost deviation data based on natural factors from the BIM database.

[0008] Based on the design cost data and the cost deviation data, engineering cost data is generated, and the engineering cost data and the geometric attribute data are correlated and calculated to generate cost geometric correlation information;

[0009] Geometric analysis is performed on the three-dimensional point cloud data to extract the three-dimensional deformation data of the component;

[0010] Based on the BIM model, the spatial connection relationships between multiple components are analyzed, and a component relationship diagram is constructed.

[0011] The three-dimensional deformation data and the cost geometric correlation information are jointly mapped to the component correlation diagram. A fusion feature map is formed using a dual-channel convolutional neural network. Data analysis is performed on the fusion feature map to generate a cost deviation probability distribution.

[0012] Based on the cost deviation probability distribution, a reinforcement learning algorithm is used to generate a cost compensation strategy that matches the current construction status. The cost compensation strategy is then automatically transmitted and updated to the BIM database to achieve dynamic adjustment of the project cost data.

[0013] Optionally, the step of generating a cost compensation strategy that matches the current construction status using a reinforcement learning algorithm based on the cost deviation probability distribution, and automatically transmitting and updating the cost compensation strategy to the BIM database, includes:

[0014] Based on the cost deviation probability distribution, the expected benefits of different cost adjustment actions are evaluated using a reinforcement learning algorithm. Based on the expected benefits, the optimal cost adjustment action is selected to generate a cost compensation strategy.

[0015] The cost compensation strategy is encapsulated in a format that meets the data structure requirements of the BIM database.

[0016] The encapsulated cost compensation strategy is written into the BIM database through a data interface, so as to automatically update the engineering cost data of relevant components based on the cost compensation strategy.

[0017] Optionally, the step of evaluating the expected benefits of different cost adjustment actions using a reinforcement learning algorithm based on the cost deviation probability distribution, and selecting the optimal cost adjustment action based on the expected benefits to generate a cost compensation strategy, includes:

[0018] The cost deviation probability distribution is used as the environmental state input to the state awareness module of the reinforcement learning algorithm;

[0019] The feature extraction layer in the state perception module is used to analyze the cost deviation characteristics and construction state characteristics in the environmental state.

[0020] The value evaluation network of the reinforcement learning algorithm calculates the expected return value of each optional cost adjustment action in the continuous state space based on the cost deviation characteristics and the construction state characteristics.

[0021] Based on the expected reward value, an action probability distribution is generated through the policy network of the reinforcement learning algorithm.

[0022] Based on the probability distribution of the actions, the cost adjustment action with the highest expected benefit is selected as the optimal cost adjustment action using the importance sampling method.

[0023] The optimal cost adjustment action is combined with component attribute data to generate a cost compensation strategy.

[0024] Optionally, the step of mapping the three-dimensional deformation data and the cost geometric correlation information together onto the component correlation graph, forming a fusion feature map using a dual-channel convolutional neural network, and performing data analysis on the fusion feature map to generate a cost deviation probability distribution includes:

[0025] The three-dimensional deformation data is mapped to the component association diagram according to the component nodes to form a deformation feature layer;

[0026] The cost geometric correlation information is mapped to the component correlation graph according to the component nodes to form a cost feature layer;

[0027] The deformation feature layer and the cost feature layer are input into a dual-channel convolutional neural network. Through the first convolutional channel of the dual-channel convolutional neural network, multi-scale feature extraction is performed on the deformation feature layer to obtain high-level deformation features under different receptive fields.

[0028] By using the second convolutional channel of the dual-channel convolutional neural network, multi-scale feature extraction is performed on the cost feature layer to obtain high-level cost features under different receptive fields;

[0029] The high-level features of deformation and the high-level features of cost are concatenated along the feature dimension through the feature fusion layer of a dual-channel convolutional neural network to generate a fused feature map.

[0030] By using the fully connected layer of a dual-channel convolutional neural network, based on historical cost deviation data, the preset overall deviation range is divided into multiple cost deviation ranges.

[0031] The probability values ​​of the fused feature map falling into each cost deviation interval are calculated through the Softmax activation function layer of a dual-channel convolutional neural network to generate a cost deviation probability distribution.

[0032] Optionally, the step of performing geometric analysis on the three-dimensional point cloud data to extract the three-dimensional deformation data of the component includes:

[0033] The three-dimensional point cloud data is divided into component units to obtain independent point cloud datasets for each component;

[0034] Noise filtering and missing data completion are performed on the independent point cloud dataset of each component. The processed point cloud data is then reconstructed to generate an actual construction surface model.

[0035] Calculate the normal offset and tangential displacement of the design geometric surface in the BIM model corresponding to each sampling point in the actual construction surface model;

[0036] Based on the normal offset and the tangential displacement, a three-dimensional deformation data field reflecting the overall deformation degree of the component is constructed;

[0037] The maximum deformation, average deformation, and deformation distribution features are extracted from the three-dimensional deformation data field to form three-dimensional deformation data.

[0038] Optionally, the step of analyzing the spatial connection relationships between multiple components based on the BIM model and constructing a component relationship diagram includes:

[0039] Analyze the connection types and spatial relative positions between components in the BIM model to establish a dataset of component connection relationships;

[0040] Based on the component connection relationship dataset, a component spatial topology network is constructed with components as nodes and spatial connection relationships between components as edges. The node features include the geometric attribute data and the cost geometric association information.

[0041] Neighborhood feature propagation is performed on the component space topology network. Node features are updated through multiple rounds of message passing. Based on the updated node features, a component association graph is constructed.

[0042] Optionally, the step of generating project cost data based on the design cost data and the cost deviation data, and performing correlation calculations on the project cost data and the geometric attribute data to generate cost geometric correlation information includes:

[0043] The design cost data and the cost deviation data are input into the cost estimation model. The multilayer perceptron in the cost estimation model performs feature fusion on the design cost data and the cost deviation data to generate engineering cost data.

[0044] Extract the component's size parameters, shape parameters, and spatial location parameters from the geometric attribute data;

[0045] A multidimensional correlation mapping is established based on the engineering cost data and the size parameters, shape parameters, and spatial location parameters to obtain the parameter correlation mapping relationship;

[0046] Based on the parameter association mapping relationship, the size parameter, the shape parameter, and the spatial position parameter are weighted and combined to generate cost geometric association information.

[0047] Secondly, this application provides a BIM-based engineering cost management system, including:

[0048] The data acquisition module is used to acquire 3D point cloud data of components from the construction site, obtain geometric attribute data of components from the BIM model, and obtain design cost data and cost deviation data based on natural factors from the BIM database.

[0049] The generation module is used to generate engineering cost data based on the design cost data and the cost deviation data, and to perform correlation calculations on the engineering cost data and the geometric attribute data to generate cost geometric correlation information.

[0050] The analysis module is used to perform geometric analysis on the three-dimensional point cloud data to extract the three-dimensional deformation data of the component;

[0051] The parsing module is used to parse the spatial connection relationships between multiple components based on the BIM model and construct a component relationship diagram;

[0052] The mapping module is used to map the three-dimensional deformation data and the cost geometric correlation information to the component correlation diagram, use a dual-channel convolutional neural network to form a fusion feature map, perform data analysis on the fusion feature map, and generate a cost deviation probability distribution.

[0053] The update module is used to generate a cost compensation strategy that matches the current construction status based on the cost deviation probability distribution using a reinforcement learning algorithm, and automatically transmits and updates the cost compensation strategy to the BIM database to achieve dynamic adjustment of the project cost data.

[0054] Thirdly, this application provides an electronic device, comprising:

[0055] Memory, used to store computer programs;

[0056] A processor, used to execute the computer program to implement the steps of the BIM-based engineering cost management method as described in the first aspect above.

[0057] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, can implement the steps of the BIM-based engineering cost management method described in the first aspect above.

[0058] This application provides a BIM-based engineering cost management method, which includes: collecting 3D point cloud data of components from the construction site; obtaining geometric attribute data of components from the BIM model; obtaining design cost data and cost deviation data based on natural factors from the BIM database; generating engineering cost data based on the design cost data and the cost deviation data; performing correlation calculations on the engineering cost data and the geometric attribute data to generate cost geometric correlation information; performing geometric analysis on the 3D point cloud data to extract 3D deformation data of the components; analyzing the spatial connection relationships between multiple components based on the BIM model to construct a component correlation diagram; mapping the 3D deformation data and the cost geometric correlation information together onto the component correlation diagram; forming a fusion feature map using a dual-channel convolutional neural network; performing data analysis on the fusion feature map to generate a cost deviation probability distribution; and using a reinforcement learning algorithm based on the cost deviation probability distribution to generate a cost compensation strategy that matches the current construction status, automatically transmitting and updating the cost compensation strategy to the BIM database to achieve dynamic adjustment of the engineering cost data.

[0059] The technical solution provided in this application has the following beneficial effects:

[0060] This application integrates on-site measured data with design data to establish a complete data foundation. It establishes a quantitative relationship between geometric properties and cost changes, accurately identifies the actual deformation state of components, reflects the spatial interaction relationships between components, predicts the possibility and extent of cost deviations, formulates cost adjustment plans adapted to on-site conditions, and achieves real-time synchronization of cost data.

[0061] Furthermore, based on the cost deviation probability distribution, this application uses a reinforcement learning algorithm to evaluate the expected benefits of different adjustment actions and select the optimal solution. After generating a cost compensation strategy, it converts the strategy into a database-compatible format and writes it into the system through an interface, automatically updating the engineering cost data of the relevant components.

[0062] Furthermore, the solution can intelligently generate the optimal compensation strategy based on real-time cost deviation prediction and automatically update the data, thereby improving the accuracy of cost adjustment and the system response speed.

[0063] These or other aspects of this application will become more apparent from the description of the following embodiments. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 A flowchart illustrating a BIM-based engineering cost management method provided in this application embodiment;

[0066] Figure 2 A schematic diagram illustrating a specific implementation of a BIM-based engineering cost management method provided in this application embodiment;

[0067] Figure 3 This is a schematic diagram of a BIM-based engineering cost management system provided in an embodiment of this application. Detailed Implementation

[0068] In building information modeling (BIM)-based project cost management, existing methods have limitations in matching cost adjustments with the actual deformation state when dealing with cost fluctuations caused by geometric deformation during construction. These limitations are mainly reflected in the insufficient consideration of the spatial correlation of components in cost prediction models and the weak adaptability of cost compensation strategies in dynamic construction environments, leading to a gap between the accuracy of cost control and the actual site conditions.

[0069] To address the aforementioned issues, this application proposes a BIM-based engineering cost management method. This method collaboratively analyzes 3D deformation data, component geometric attributes, and cost information collected on-site to establish a cost prediction model reflecting the spatial relationships between components. Specifically, the method constructs a component relationship graph to analyze deformation propagation paths, calculates cost deviation probabilities using a dual-channel neural network, and employs reinforcement learning to generate adaptive compensation strategies. This method achieves dynamic matching between engineering cost data and actual construction conditions, effectively solving the problem of cost control being disconnected from on-site conditions in existing technologies, and improving the accuracy and real-time performance of cost control.

[0070] To enable those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0071] The core of this application is to provide a BIM-based engineering cost management method, and a flowchart of one specific implementation method is shown below. Figure 1 As shown, the method includes:

[0072] Step 101: Collect 3D point cloud data of components from the construction site, obtain geometric attribute data of components from the BIM model, and obtain design cost data and cost deviation data based on natural factors from the BIM database.

[0073] In the above scheme, the BIM model refers to the digital design model corresponding to the "construction site components," which serves as a design benchmark for comparing the geometric state of the components with the actual construction components. Three-dimensional point cloud data is a set of points containing spatial location information of the component surface, obtained through laser scanning. Geometric attribute data consists of the component's design dimensions and shape parameters recorded in the building information model. Design cost data represents the planned cost of the component during the budgeting phase. Cost deviation data based on natural factors records historical data on cost changes caused by environmental factors such as climate conditions.

[0074] In this embodiment, firstly, a laser scanning equipment array is used to collect the spatial coordinates of the surface of the components at the construction site to form three-dimensional point cloud data. For example, six scanners are deployed in a commercial complex project to collect data from concrete beams and columns from multiple angles. Secondly, the cross-sectional dimensions and length parameters of the beam components are retrieved from the building information model server as geometric attribute data, such as extracting the width, height, and span data of the main beam cross-section. Then, the original budget value of the component is read from the cost management database as design cost data, and records of additional expenditures caused by recent rainy season construction are retrieved as cost deviation data, such as obtaining the design cost of the beam components and the additional equipment rental costs incurred due to work stoppages caused by rainfall.

[0075] In practical applications, a convention center project acquired point cloud data of steel roof trusses through ground-based fixed scanning stations, simultaneously extracted roof truss member specification information from the model library, and obtained initial cost and repair cost records due to typhoon impact from the financial system, thus completing multi-source data collection.

[0076] This application integrates on-site measured data with design management data to establish a complete and accurate data foundation for subsequent analysis.

[0077] Step 102: Based on the design cost data and the cost deviation data, generate engineering cost data, perform correlation calculations on the engineering cost data and the geometric attribute data, and generate cost geometric correlation information.

[0078] In this embodiment, step 102 includes the following process:

[0079] Step 1021: Input the design cost data and the cost deviation data into the cost estimation model, and perform feature fusion on the design cost data and the cost deviation data through the multilayer perceptron in the cost estimation model to generate engineering cost data.

[0080] Step 1022: Extract the component's size parameters, shape parameters, and spatial location parameters from the geometric attribute data.

[0081] Step 1023: Establish a multi-dimensional correlation mapping based on the engineering cost data and the size parameters, shape parameters, and spatial location parameters to obtain the parameter correlation mapping relationship.

[0082] Step 1024: Based on the parameter association mapping relationship, the size parameter, the shape parameter, and the spatial position parameter are weighted and combined to generate cost geometric association information.

[0083] In the above scheme, the project cost data is a dynamic cost estimate generated by integrating the design cost and historical deviations. The cost geometric correlation information is a data set that establishes a quantitative relationship between project cost and component geometric parameters. Dimensional parameters describe the measured values ​​of the component's length, width, and height; shape parameters characterize the component's cross-sectional morphology; and spatial position parameters record the component's coordinates and orientation in three-dimensional space. The parameter correlation mapping relationship is a data structure that reveals the degree of influence of different geometric parameters on cost.

[0084] In this embodiment, firstly, in step 1021, design cost data and cost deviation data are input into a cost estimation model containing multiple calculation nodes. This model integrates the two types of cost features through layer-by-layer weighted calculation, and outputs engineering cost data reflecting the current construction conditions. For example, the design cost of a concrete beam and the deviation data caused by rainy season construction are input into the model, and a dynamic cost including the influence of environmental factors is obtained through a three-layer neural network calculation. Secondly, in step 1022, dimensional parameters such as the length, width, and height of the components, shape parameters such as the cross-sectional shape coefficient, and spatial location parameters such as elevation coordinates are separated from the geometric attribute data. For example, the span value, cross-sectional rectangularity, and installation height data of the beam components are extracted.

[0085] Next, step 1023 involves performing a multidimensional correlation analysis between the project cost data and the three types of geometric parameters to establish the correspondence between parameter changes and cost fluctuations. For example, it analyzes the cost growth pattern for each unit increase in beam height. Finally, step 1024 determines the weight ratio of each geometric parameter based on the parameter correlation mapping relationship, and performs weighted fusion of dimensional parameters, shape parameters, and spatial location parameters to generate cost geometric correlation information. For example, the beam height parameter is assigned a higher weight and combined with other parameters to form a cost influence coefficient.

[0086] In practical application, taking into account the data collection results of the steel roof truss of the aforementioned convention center project, this embodiment first inputs the steel roof truss design cost data of 85,000 yuan and the cost deviation data of 12,000 yuan caused by the typhoon into the cost estimation model. The multilayer perception network in this model contains three calculation layers. Through weighted calculation, the two types of cost data are merged to generate the project cost data as 85,000×0.7+12,000×0.3=59,500+3,600=63,100 yuan, where the weight coefficients are obtained through model training. Then, the dimensional parameters are extracted from the geometric attribute data of the steel roof truss, including the member length of 6.5 meters, the cross-sectional width of 0.3 meters, and the height of 0.4 meters. The shape parameter adopts a rectangular cross-sectional coefficient of 0.92. The spatial position parameters include the installation height of 12.5 meters and the horizontal coordinates [35.2, 18.6]. A multidimensional correlation mapping was established between the project cost data of 63,100 yuan and these parameters. Regression analysis revealed the following correlation: for every 0.01 meter increase in cross-sectional height, the cost increases by 85 yuan; for every 1 meter increase in installation height, the cost increases by 120 yuan. The correlation degree was calculated using the Pearson correlation coefficient formula. in This represents the Pearson correlation coefficient. For parameter changes, This represents the cost change. Based on this mapping relationship, the size parameter is assigned a weight of 0.38, the shape parameter a weight of 0.25, and the spatial location parameter a weight of 0.40. The weight allocation is determined according to the correlation strength between each parameter and cost. Finally, the cost geometric correlation information is generated through the weighted combination formula: 0.38 × size parameter influence value + 0.25 × shape parameter influence value + 0.40 × spatial location parameter influence value. This information clearly records the cost impact corresponding to a unit change in each geometric parameter.

[0087] This application provides a quantitative basis for cost prediction under deformation conditions by establishing a dynamic correlation model between cost and geometric features.

[0088] Step 103: Perform geometric analysis on the three-dimensional point cloud data to extract the three-dimensional deformation data of the component.

[0089] In this embodiment, step 103 includes the following process:

[0090] Step 1031: Divide the three-dimensional point cloud data into component units to obtain independent point cloud datasets for each component.

[0091] Step 1032: Perform noise filtering and missing data completion processing on the independent point cloud dataset of each component, and reconstruct the surface of the processed point cloud data to generate the actual construction surface model.

[0092] Step 1033: Calculate the normal offset and tangential displacement of the design geometric surface in the BIM model corresponding to each sampling point in the actual construction surface model.

[0093] Step 1034: Construct a three-dimensional deformation data field that reflects the overall deformation degree of the component based on the normal offset and the tangential displacement.

[0094] Step 1035: Extract the maximum deformation, average deformation, and deformation distribution features from the three-dimensional deformation data field to form three-dimensional deformation data.

[0095] In the above scheme, a component unit refers to an independent functional part that constitutes a building. Its classification is based on a predefined component classification system in the BIM model, determined by identifying point sets with continuous geometric features in the point cloud data and matching them with the components in the BIM model. The design geometric surfaces in the BIM model are components of the geometric attribute data; the two have an inclusion relationship. In addition to the design geometric surface information, the geometric attribute data also includes the component's dimensional parameters, material properties, and other geometric feature data. An independent point cloud dataset is a collection of spatial points belonging to a single component after segmentation. The actual construction surface model is the real surface model of the component obtained through point cloud reconstruction. The normal offset is the displacement of the actual surface point along the normal direction of the design surface, and the tangential displacement is the distance the actual surface point moves within the tangential plane of the design surface. The three-dimensional deformation data field records the spatial distribution data of the deformation values ​​at various locations of the component; the deformation distribution characteristics describe the variation law of the deformation on the component surface. Three-dimensional deformation data is a set of quantitative data that characterizes the geometric differences between the actual construction state and the design state of a component. It includes spatial displacement information at each location of the component, overall deformation statistical characteristics, and deformation spatial distribution patterns, and is used to accurately describe the dimensional deviations and shape changes that occur during construction.

[0096] In this embodiment, step 1031 first segments the 3D point cloud data based on the spatial range of the components, aggregating point clouds belonging to the same component into independent point cloud datasets. For example, the point cloud of the steel roof truss of the convention center is separated into point cloud sets of chord members and web members according to truss member units. The specific implementation process of "segmenting the 3D point cloud data according to component units" is as follows: First, based on the predefined component classification and spatial range in the BIM model, corresponding spatial boundaries are set in the point cloud data. Then, a density clustering algorithm is used to identify point cloud sets with continuous distribution characteristics within each boundary. Finally, point clouds belonging to the same component are aggregated into independent datasets. For example, when processing the point cloud of a concrete column, the system first locates the design position of the column, and then identifies the area with a point cloud density higher than a set threshold within the spatial range as the actual point cloud of the column, thereby separating the independent point cloud dataset of the column.

[0097] Secondly, in step 1032, noise removal and data completion are performed on the point cloud of each component. A continuous actual construction surface model is generated using a surface fitting algorithm. For example, after denoising the point cloud of the steel truss chord, the actual member surface is generated by polynomial surface fitting. The specific implementation process of "reconstructing the surface of the processed point cloud data to generate the actual construction surface model" is as follows: the moving least squares method is used to perform local surface fitting on the denoised and completed point cloud. The best fitted surface patch in each neighborhood is determined through iterative calculation. Finally, all surface patches are spliced ​​together to form a complete surface model. For example, when reconstructing the surface of a beam component, the algorithm will select neighboring points within a 5 cm radius around each sampling point to calculate the least squares fitting plane, and then merge all the small smooth surface patches to generate the actual construction surface model of the beam.

[0098] Next, in step 1033, the actual constructed surface is registered with the design model surface, and the offset of each sampling point in the normal direction and the displacement in the tangential direction of the design surface are calculated. For example, the vertical settlement and horizontal offset of the chord surface points relative to the design position are calculated. Then, in step 1034, a three-dimensional deformation data field is constructed based on the normal offset and tangential displacement of all sampling points to form a spatial distribution map of component deformation. For example, a grid distribution of deformation values ​​at various locations of the chord is established. Finally, in step 1035, the maximum and average deformation values ​​are statistically analyzed from the three-dimensional deformation data field to analyze the deformation distribution characteristics. For example, the characteristic of larger deformation in the middle and smaller deformation at both ends of the chord is identified, forming complete three-dimensional deformation data.

[0099] In practical application, taking into account the cost geometric correlation information generation results of the steel roof truss of the aforementioned convention center project, this embodiment first divides the collected 3D point cloud data of the steel roof truss according to truss member units to obtain independent point cloud datasets for chords and web members, where the main chord point cloud contains 12,560 spatial points. After noise filtering to remove 83 outliers from the chord point cloud, missing areas are filled in by neighbor interpolation. The moving least squares method is used to reconstruct the surface and generate an actual construction surface model. This model consists of 2,512 surface patches with a reconstruction accuracy of 0.005 meters. After registering the actual construction surface model with the design geometric surface in the building information model, the normal offset and tangential displacement of each sampling point corresponding to the design surface are calculated. The maximum normal offset is 0.008 meters and the maximum tangential displacement is 0.005 meters. The normal offset is calculated using the vector projection formula. ,in Indicates the normal offset. These are the actual point coordinates. To design the coordinates of the corresponding points on the curved surface, To design the unit normal vector of the curved surface, a three-dimensional deformation data field was constructed based on the normal offset and tangential displacement of all sampling points, forming a deformation value distribution grid covering the entire surface of the chord with a grid size of 0.01 m × 0.01 m. The maximum deformation of 0.008 m was extracted from this three-dimensional deformation data field, and the average deformation was calculated to be 0.0032 m using an arithmetic mean. Simultaneously, the deformation distribution characteristics were identified, showing that the deformation was larger in the middle of the chord and smaller at both ends, ultimately forming three-dimensional deformation data containing specific deformation values ​​and distribution patterns.

[0100] This application provides an accurate basis for cost deviation prediction by precisely analyzing the geometric deformation state of components.

[0101] Step 104: Based on the BIM model, analyze the spatial connection relationship between multiple components and construct a component association diagram.

[0102] In this embodiment, step 104 includes the following process:

[0103] Step 1041: Analyze the connection types and spatial relative positions between components in the BIM model to establish a dataset of component connection relationships.

[0104] Step 1042: Based on the component connection relationship dataset, construct a component spatial topology network with components as nodes and spatial connection relationships between components as edges, wherein the node features include the geometric attribute data and the cost geometric association information.

[0105] Step 1043: Propagate neighborhood features on the component space topology network, update node features through multiple rounds of message passing, and construct a component association graph based on the updated node features.

[0106] In the above scheme, connection type and spatial relative position jointly define the specific content of "spatial connection relationship". The connection type describes how components are connected (e.g., hinged, rigid), and the spatial relative position describes the orientation relationship of components in space (e.g., adjacent, supported). The combination of the two fully characterizes the topological connection characteristics between components. The component connection relationship dataset is a collection of data recording the connection methods and spatial orientations between components. The component spatial topology network is a graph structure composed of components as nodes and connection relationships as edges. Neighborhood feature propagation is the process of transmitting and aggregating feature information through the connection relationships between nodes. The component association graph is an enhanced graph model that includes the mutual influence relationships between components.

[0107] In this embodiment, firstly, step 1041 analyzes the connection types between components in the Building Information Model (BIM), including hinged and rigid connections, and the spatial relative positions, including adjacent and supported relationships, to establish a component connection relationship dataset. For example, it analyzes the welding connection types and spatial intersection relationships between chords and web members in the steel roof truss of an exhibition center. Secondly, step 1042, based on the component connection relationship dataset, constructs a component spatial topology network with each member as a node and the connection relationships between members as edges. Geometric attribute data, including member dimensions and cost geometric correlation information, including cost impact coefficients, are assigned to the corresponding nodes. For example, the cost impact value of 85 yuan for the change in the length and cross-sectional dimensions of the chord is used as a node feature.

[0108] Next, in step 1043, three rounds of neighborhood feature propagation are performed on the component space topology network. In each round, the features of adjacent nodes are aggregated and the current node features are updated through a message passing mechanism. Based on the updated node features, a component association graph is constructed. For example, after the chord node receives features from the adjacent web member node, it updates its own feature vector, and finally forms an association graph that reflects the cost propagation relationship between members.

[0109] In practical applications, this embodiment utilizes the 3D deformation data generation results of the steel roof truss for the aforementioned convention center project. First, it analyzes the connection types and spatial relative positions between the steel roof truss components in the Building Information Model (BIM) to establish a component connection relationship dataset. The chord and web members are connected by welding with a connection strength coefficient of 0.9. The spatial relative positions are recorded as the coordinates of the intersection points of the chord and web members [12.3, 8.7, 15.6]. Based on this dataset, a component spatial topology network is constructed, with 16 nodes representing all members. 24 edges between nodes are established according to the actual connection relationships. Geometric attribute data, including the chord cross-sectional dimensions of 0.3m × 0.4m and cost geometric correlation information, including the cost impact value of 85 yuan for changes in cross-sectional height, are assigned to the corresponding nodes. Three rounds of neighborhood feature propagation are performed on the component spatial topology network, updating node features through a message passing mechanism. The node feature update uses a graph convolution formula. ,in Indicates the first Layer node feature matrix This represents the activation function. Indicates the first Layer weight matrix, Indicates the first Layer adjacent nodes eigenvectors, Indicates the first Layer bias term. After three rounds of iteration, the update characteristics of the chord node show that its average cost impact coefficient on adjacent components reaches 0.76. This value is obtained by weighting the feature values ​​passed by each connection edge. The final component relationship diagram fully represents the cost relationship between the roof truss members.

[0110] This application accurately characterizes the propagation mechanism of the cost impact of spatial interactions between components by constructing a component relationship diagram.

[0111] Step 105: Map the three-dimensional deformation data and the cost geometric correlation information together to the component correlation diagram, use a dual-channel convolutional neural network to form a fusion feature map, perform data analysis on the fusion feature map, and generate a cost deviation probability distribution.

[0112] In this embodiment, step 105 includes the following process:

[0113] Step 1051: Map the three-dimensional deformation data to the component association diagram according to the component nodes to form a deformation feature layer.

[0114] Step 1052: Map the cost geometric association information to the component association graph according to the component nodes to form a cost feature layer.

[0115] Step 1053: Input the deformation feature layer and the cost feature layer into a dual-channel convolutional neural network. Through the first convolutional channel of the dual-channel convolutional neural network, perform multi-scale feature extraction on the deformation feature layer to obtain high-level deformation features under different receptive fields.

[0116] Step 1054: Through the second convolutional channel of the dual-channel convolutional neural network, perform multi-scale feature extraction on the cost feature layer to obtain high-level cost features under different receptive fields.

[0117] Step 1055: Using the feature fusion layer of a dual-channel convolutional neural network, the deformation high-level features and the cost high-level features are concatenated along the feature dimension to generate a fused feature map.

[0118] Step 1056: Using the fully connected layer of a dual-channel convolutional neural network, based on historical cost deviation data, divide the preset overall deviation range into multiple cost deviation ranges.

[0119] Step 1057: Calculate the probability value of the fused feature map falling into each cost deviation interval through the Softmax activation function layer of the dual-channel convolutional neural network to generate a cost deviation probability distribution.

[0120] In the above scheme, historical cost deviation data is obtained from the BIM engineering database and records the differences between the actual cost and design cost of components caused by various factors (including but not limited to natural factors, construction factors, etc.) in past projects. It includes specific deviation values ​​and corresponding occurrence conditions. The deformation feature layer is a data layer recording the deformation state of each node in the component relationship diagram. The cost feature layer is a data layer recording the cost impact characteristics of each node. Advanced deformation features are deformation spatial pattern features extracted through multi-scale convolution, and advanced cost features are cost spatial pattern features extracted through multi-scale convolution. The fusion feature map is a comprehensive data map integrating deformation and cost features. The cost deviation range is a cost deviation level divided based on historical data. The cost deviation probability distribution describes the probability distribution of different levels of cost deviation for components in future construction stages. It quantifies the probability of a component falling into each preset cost deviation range using probability values, and is used to predict and assess cost risks.

[0121] In this embodiment, firstly, step 1051 maps the three-dimensional deformation data to the component relationship diagram according to the component nodes to form a deformation feature layer. For example, the deformation of 0.008 meters of the steel roof truss chord of the convention center is assigned to the corresponding node. Secondly, step 1052 maps the cost geometric relationship information to the component relationship diagram according to the component nodes to form a cost feature layer. For example, the cost impact value of 85 yuan for the change in the chord cross-section height is assigned to the corresponding node. Then, step 1053 inputs the deformation feature layer and the cost feature layer into a dual-channel convolutional neural network. The first convolutional channel uses three different sizes of convolutional kernels to perform multi-scale feature extraction on the deformation feature layer to obtain high-level deformation features under different receptive fields, such as extracting local deformation details and overall deformation trend features of the chord.

[0122] Subsequently, step 1054 involves multi-scale feature extraction through the second convolutional channel of the cost feature layer to obtain high-level cost features under different receptive fields, such as extracting local fluctuations and global distribution features of the cost impact on the chord. Then, step 1055 uses a feature fusion layer to concatenate the high-level deformation features and cost features along the feature dimension, generating a fused feature map, for example, combining the deformation features of the chord with the cost features into a comprehensive feature vector. Next, step 1056 uses a fully connected layer to divide the preset overall deviation range into multiple cost deviation ranges based on historical cost deviation data, for example, dividing the historical cost data of the convention center into three ranges: 0-50 yuan, 50-100 yuan, and above 100 yuan. Finally, step 1057 uses a Softmax activation function layer to calculate the probability value of the fused feature map falling into each cost deviation range, generating a cost deviation probability distribution, for example, calculating the probability that the chord belongs to the three cost deviation ranges in the current state as 25%, 45%, and 30%, respectively.

[0123] In practical application, building upon the component relationship diagram construction results of the aforementioned convention center project's steel roof truss, this embodiment first maps the three-dimensional deformation data, including the maximum deformation of the chord members (0.008 meters) and the average deformation (0.0032 meters), to the component relationship diagram according to the component nodes, forming a deformation feature layer. The chord member nodes record their actual deformation state. Secondly, the cost geometric relationship information, including the cost impact value of 85 yuan due to changes in cross-sectional height, is mapped to the component relationship diagram according to the component nodes, forming a cost feature layer. The chord member nodes record their cost sensitivity. The deformation feature layer and the cost feature layer are input into a dual-channel convolutional neural network. The first convolutional channel uses three different kernel sizes (3×3, 5×5, and 7×7) to perform multi-scale feature extraction on the deformation feature layer, obtaining high-level deformation features under different receptive fields, such as extracting local subtle deformation and overall bending features of the chord members. The second convolutional channel also uses multi-scale kernels to extract features from the cost feature layer, obtaining high-level cost features under different receptive fields, such as extracting local abrupt changes and global trend features of the cost impact of the chord members. The feature fusion layer concatenates high-level deformation features and high-level cost features along the feature dimension to generate a fused feature map. For example, the 24-dimensional deformation features of the chord are concatenated with the 18-dimensional cost features to form a 42-dimensional comprehensive feature vector. The fully connected layer, based on the historical cost deviation data of the convention center over the past 12 months, divides the preset overall deviation range of 0-200 yuan into three cost deviation ranges: 0-50 yuan, 50-100 yuan, and 100-200 yuan. The range division is determined based on the distribution density of historical data. The Softmax activation function layer calculates the probability value of the fused feature map falling into each cost deviation range, generating a cost deviation probability distribution. The probability calculation uses the Softmax formula. ,in For the first The original scores for each interval, For the first The probability values ​​for each interval. This represents the number of cost deviation intervals. The calculated probability values ​​for the chord belonging to the three intervals are 0.25, 0.45, and 0.30, respectively, and the sum of these probability values ​​is 1.

[0124] This application achieves accurate prediction of cost deviation probability through multi-scale analysis that integrates deformation and cost characteristics.

[0125] Step 106: Based on the cost deviation probability distribution, a reinforcement learning algorithm is used to generate a cost compensation strategy that matches the current construction status. The cost compensation strategy is automatically transmitted and updated to the BIM database to achieve dynamic adjustment of the project cost data.

[0126] In this embodiment, step 106 includes the following process, such as... Figure 2 As shown:

[0127] Step 1061: Based on the cost deviation probability distribution, evaluate the expected benefits of different cost adjustment actions using a reinforcement learning algorithm, and select the optimal cost adjustment action based on the expected benefits to generate a cost compensation strategy.

[0128] Step 1061 may specifically include the following steps:

[0129] A1: The cost deviation probability distribution is used as the environmental state input to the state awareness module of the reinforcement learning algorithm.

[0130] A2: The feature extraction layer in the state perception module is used to analyze the cost deviation features and construction state features in the environmental state.

[0131] A3: Using the value evaluation network of the reinforcement learning algorithm, based on the cost deviation characteristics and the construction state characteristics, calculate the expected revenue value of each optional cost adjustment action in the continuous state space.

[0132] A4: The policy network of the reinforcement learning algorithm generates an action probability distribution based on the expected reward value.

[0133] A5: Based on the probability distribution of the actions, the cost adjustment action with the highest expected benefit is selected as the optimal cost adjustment action using the importance sampling method.

[0134] A6: Combine the optimal cost adjustment action with component attribute data to generate a cost compensation strategy.

[0135] Step 1062: Encapsulate the cost compensation strategy in a format that meets the data structure requirements of the BIM database.

[0136] Step 1063: Write the encapsulated cost compensation strategy into the BIM database through the data interface, so as to automatically update the engineering cost data of relevant components based on the cost compensation strategy.

[0137] In the above scheme, the current construction status is obtained through the fusion analysis of three-dimensional deformation data and cost geometric relationship information, reflecting the real-time correlation between the actual geometric state of the component and the cost impact. The application of the reinforcement learning algorithm in this application differs from existing technologies in that: existing technologies typically directly select the action with the highest expected return value, while this method constructs an action probability distribution and employs importance sampling to ensure the selection of high-return actions while also exploring other potential optimization solutions, thereby avoiding getting trapped in local optima. A continuous state space refers to a state representation composed of continuously changing probability values ​​in the cost deviation probability distribution. Its continuity is reflected in the fact that the state feature values ​​are continuous values ​​in the real number field, rather than discrete finite state points. The application of the importance sampling method in this step adjusts the sampling weight of each action by adjusting its selection probability in the action probability distribution, so that actions with high expected returns have a higher selection probability without completely excluding other actions. Its effect is to prioritize the current optimal action while retaining the opportunity to explore other potentially better actions, thus improving the robustness and adaptability of the strategy. Component attribute data refers to a set of data describing the basic characteristics of components, sourced from the BIM model database. It is inclusive of geometric attribute data, which specifically describes the geometric features of the component. Cost compensation strategies are specific adjustment schemes proposed to address cost deviations. The state awareness module is a component in reinforcement learning algorithms that analyzes environmental states. The action probability distribution is the set of probabilities that each available action will be selected.

[0138] Cost deviation features are key indicators extracted from the cost deviation probability distribution, describing the degree and distribution pattern of the deviation and describing the specific situation of the current cost deviating from the expected state. Construction status features are parameters extracted from on-site monitoring data, reflecting construction progress, environmental conditions, and resource status, characterizing the real-time state of the current construction environment. Expected return value is a quantitative assessment of the comprehensive future benefits that each optional cost adjustment action may bring, including a comprehensive consideration of cost savings and execution difficulty. Action probability distribution is the probability distribution of each optional cost adjustment action output by the reinforcement learning policy network, reflecting the relative advantage of each action in the current state. The optimal cost adjustment action is the cost adjustment scheme with the highest expected return value selected from the action probability distribution using an importance sampling method; this scheme best meets the cost control requirements of the current construction state.

[0139] In this embodiment, firstly, the cost deviation probability distribution is used as the environmental state input to the state perception module in step 1061. The feature extraction layer of this module analyzes the cost deviation features and construction state features in the environmental state. The value assessment network calculates the expected benefit value of each cost adjustment action based on these features. The strategy network generates an action probability distribution based on the expected benefit value, selects the cost adjustment action with the highest expected benefit using an importance sampling method, and finally combines the action with the component attribute data to generate a cost compensation strategy. For example, after inputting the cost deviation probability distribution of the steel roof truss of the convention center, after feature analysis and benefit assessment, the material usage adjustment is selected as the optimal action, and a specific compensation scheme is generated in combination with the chord specifications.

[0140] Secondly, in step 1062, the cost compensation strategy is encapsulated according to the database structure requirements, such as converting the monetary values ​​and execution conditions in the compensation strategy into the corresponding database field formats. Finally, in step 1063, the encapsulated strategy is written into the database through a data interface, automatically updating the engineering cost data of relevant components. For example, the cost compensation data of the chord is written into the building information model database through an application programming interface, updating the cost information of the component in real time.

[0141] In practical applications, this embodiment utilizes the cost deviation probability distribution generated from the steel roof truss of the aforementioned convention center project. Firstly, the cost deviation probability distribution is used as the environmental state input to the state perception module of the reinforcement learning algorithm. The feature extraction layer of this module analyzes the cost deviation features in the environmental state, including probability values ​​of 0.25, 0.45, and 0.30 in three intervals, and construction state features including current construction progress of 85%, ambient temperature of 28 degrees Celsius, and humidity of 70%. Based on these features, a value assessment network calculates the expected revenue values ​​of three optional cost adjustment actions: adjusting the material ratio yields a revenue of -15 points, increasing the number of construction personnel yields a revenue of 32 points, and optimizing the construction sequence yields a revenue of 78 points. The revenue value calculation uses the Q-value update formula. ,in Representing state Next action Q value, Indicates the current environmental state. This indicates an optional cost adjustment action. The learning rate is set to 0.1. This indicates all optional cost adjustment actions. The discount factor is set to 0.9. The immediate reward is determined based on the cost of the action. The next state is defined as follows. The strategy network generates action probability distributions based on expected return values: adjusting material mix ratios by 20%, increasing construction personnel by 35%, and optimizing construction sequence by 45%. Using an importance sampling method, the action with the highest expected return value is selected as the optimal cost adjustment action. This action is combined with the chord member attribute data, including concrete strength grade C30 and curing period of 7 days, to generate a cost compensation strategy. Specific strategies include adjusting the construction sequence to nighttime pouring, extending vibration time by 15%, and increasing the frequency of quality inspections. The cost compensation strategy is encapsulated according to the JSON format requirements of the Building Information Modeling (BIM) database, forming a standard data package containing a compensation amount of 82 yuan, the concrete pouring period, and the applicable chord member. Finally, the encapsulated cost compensation strategy is written to the BIM database via a RESTful application programming interface. The system automatically updates the chord member's project cost data from 63,100 yuan to 63,182 yuan, completing the dynamic adjustment of the cost data.

[0142] This application achieves dynamic matching and precise control of project cost and construction status by intelligently generating cost compensation strategies and automatically updating data.

[0143] Figure 3 A schematic diagram of a BIM-based engineering cost management system provided in this application embodiment is shown below. Figure 3 As shown, the detailed implementation section describes:

[0144] The acquisition module 31 is used to acquire three-dimensional point cloud data of components from the construction site, obtain geometric attribute data of components from the BIM model, and obtain design cost data and cost deviation data based on natural factors of components from the BIM database.

[0145] The generation module 32 is used to generate engineering cost data based on the design cost data and the cost deviation data, and to perform correlation calculations on the engineering cost data and the geometric attribute data to generate cost geometric correlation information.

[0146] Analysis module 33 is used to perform geometric analysis on the three-dimensional point cloud data to extract the three-dimensional deformation data of the component.

[0147] The parsing module 34 is used to parse the spatial connection relationship between multiple components based on the BIM model and construct a component relationship diagram.

[0148] The mapping module 35 is used to map the three-dimensional deformation data and the cost geometric correlation information to the component correlation diagram, use a dual-channel convolutional neural network to form a fusion feature map, perform data analysis on the fusion feature map, and generate a cost deviation probability distribution.

[0149] The update module 36 is used to generate a cost compensation strategy that matches the current construction status based on the cost deviation probability distribution using a reinforcement learning algorithm, and automatically transmit and update the cost compensation strategy to the BIM database to achieve dynamic adjustment of the project cost data.

[0150] The BIM-based engineering cost management system of this application embodiment is used to implement the aforementioned BIM-based engineering cost management method. Therefore, the specific implementation of the BIM-based engineering cost management system can be found in the embodiment section of the BIM-based engineering cost management method above. The specific implementation can be referred to the description of the corresponding embodiments, which will not be repeated here.

[0151] This application also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of any of the BIM-based engineering cost management methods described above.

[0152] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the BIM-based engineering cost management methods described above.

[0153] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0154] The embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the BIM-based engineering cost management method.

[0155] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0156] The foregoing has provided a detailed description of a BIM-based engineering cost management method, system, electronic device, and storage medium provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and its core ideas. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of this application.

Claims

1. A BIM-based engineering cost management method, characterized in that, include: The system collects 3D point cloud data of components from the construction site, obtains geometric attribute data of components from the BIM model, and obtains design cost data and cost deviation data based on natural factors from the BIM database. Based on the design cost data and the cost deviation data, engineering cost data is generated, and the engineering cost data and the geometric attribute data are correlated and calculated to generate cost geometric correlation information; Geometric analysis is performed on the three-dimensional point cloud data to extract the three-dimensional deformation data of the component; Based on the BIM model, the spatial connection relationships between multiple components are analyzed, and a component relationship diagram is constructed. The three-dimensional deformation data and the cost geometric correlation information are jointly mapped to the component correlation diagram. A fusion feature map is formed using a dual-channel convolutional neural network. Data analysis is performed on the fusion feature map to generate a cost deviation probability distribution. Based on the cost deviation probability distribution, a reinforcement learning algorithm is used to generate a cost compensation strategy that matches the current construction status. The cost compensation strategy is automatically transmitted and updated to the BIM database to achieve dynamic adjustment of the project cost data. The step of generating a cost compensation strategy that matches the current construction status using a reinforcement learning algorithm based on the cost deviation probability distribution, and automatically transmitting and updating the cost compensation strategy to the BIM database, includes: Based on the cost deviation probability distribution, the expected benefits of different cost adjustment actions are evaluated using a reinforcement learning algorithm. Based on the expected benefits, the optimal cost adjustment action is selected to generate a cost compensation strategy. The cost compensation strategy is encapsulated in a format that meets the data structure requirements of the BIM database. The encapsulated cost compensation strategy is written into the BIM database through a data interface, so as to automatically update the engineering cost data of relevant components based on the cost compensation strategy. The process of mapping the three-dimensional deformation data and the cost geometric correlation information together onto the component correlation graph, forming a fused feature map using a dual-channel convolutional neural network, and performing data analysis on the fused feature map to generate a cost deviation probability distribution includes: The three-dimensional deformation data is mapped to the component association diagram according to the component nodes to form a deformation feature layer; The cost geometric correlation information is mapped to the component correlation graph according to the component nodes to form a cost feature layer; The deformation feature layer and the cost feature layer are input into a dual-channel convolutional neural network. Through the first convolutional channel of the dual-channel convolutional neural network, multi-scale feature extraction is performed on the deformation feature layer to obtain high-level deformation features under different receptive fields. By using the second convolutional channel of the dual-channel convolutional neural network, multi-scale feature extraction is performed on the cost feature layer to obtain high-level cost features under different receptive fields; The high-level features of deformation and the high-level features of cost are concatenated along the feature dimension through the feature fusion layer of a dual-channel convolutional neural network to generate a fused feature map. By using the fully connected layer of a dual-channel convolutional neural network, based on historical cost deviation data, the preset overall deviation range is divided into multiple cost deviation ranges. The probability values ​​of the fused feature map falling into each cost deviation interval are calculated through the Softmax activation function layer of a dual-channel convolutional neural network to generate a cost deviation probability distribution.

2. The method according to claim 1, characterized in that, The step of evaluating the expected benefits of different cost adjustment actions using a reinforcement learning algorithm based on the cost deviation probability distribution, and selecting the optimal cost adjustment action based on the expected benefits to generate a cost compensation strategy, includes: The cost deviation probability distribution is used as the environmental state input to the state awareness module of the reinforcement learning algorithm; The feature extraction layer in the state perception module is used to analyze the cost deviation characteristics and construction state characteristics in the environmental state. The value evaluation network of the reinforcement learning algorithm calculates the expected return value of each optional cost adjustment action in the continuous state space based on the cost deviation characteristics and the construction state characteristics. Based on the expected reward value, an action probability distribution is generated through the policy network of the reinforcement learning algorithm. Based on the probability distribution of the actions, the cost adjustment action with the highest expected benefit is selected as the optimal cost adjustment action using the importance sampling method. The optimal cost adjustment action is combined with component attribute data to generate a cost compensation strategy.

3. The method according to claim 1, characterized in that, The geometric analysis of the three-dimensional point cloud data to extract the three-dimensional deformation data of the component includes: The three-dimensional point cloud data is divided into component units to obtain independent point cloud datasets for each component; Noise filtering and missing data completion are performed on the independent point cloud dataset of each component. The processed point cloud data is then reconstructed to generate an actual construction surface model. Calculate the normal offset and tangential displacement of the design geometric surface in the BIM model corresponding to each sampling point in the actual construction surface model; Based on the normal offset and the tangential displacement, a three-dimensional deformation data field reflecting the overall deformation degree of the component is constructed; The maximum deformation, average deformation, and deformation distribution features are extracted from the three-dimensional deformation data field to form three-dimensional deformation data.

4. The method according to claim 1, characterized in that, The process of analyzing the spatial connection relationships between multiple components based on the BIM model and constructing a component relationship diagram includes: Analyze the connection types and spatial relative positions between components in the BIM model to establish a dataset of component connection relationships; Based on the component connection relationship dataset, a component spatial topology network is constructed with components as nodes and spatial connection relationships between components as edges. The node features include the geometric attribute data and the cost geometric association information. Neighborhood feature propagation is performed on the component space topology network. Node features are updated through multiple rounds of message passing. Based on the updated node features, a component association graph is constructed.

5. The method according to claim 1, characterized in that, The process involves generating project cost data based on the design cost data and the cost deviation data, and performing correlation calculations between the project cost data and the geometric attribute data to generate cost geometric correlation information, including: The design cost data and the cost deviation data are input into the cost estimation model. The multilayer perceptron in the cost estimation model performs feature fusion on the design cost data and the cost deviation data to generate engineering cost data. Extract the component's size parameters, shape parameters, and spatial location parameters from the geometric attribute data; A multidimensional correlation mapping is established based on the engineering cost data and the size parameters, shape parameters, and spatial location parameters to obtain the parameter correlation mapping relationship; Based on the parameter association mapping relationship, the size parameter, the shape parameter, and the spatial position parameter are weighted and combined to generate cost geometric association information.

6. A BIM-based engineering cost management system, characterized in that, include: The data acquisition module is used to acquire 3D point cloud data of components from the construction site, obtain geometric attribute data of components from the BIM model, and obtain design cost data and cost deviation data based on natural factors from the BIM database. The generation module is used to generate engineering cost data based on the design cost data and the cost deviation data, and to perform correlation calculations on the engineering cost data and the geometric attribute data to generate cost geometric correlation information. The analysis module is used to perform geometric analysis on the three-dimensional point cloud data to extract the three-dimensional deformation data of the component; The parsing module is used to parse the spatial connection relationships between multiple components based on the BIM model and construct a component relationship diagram; The mapping module is used to map the three-dimensional deformation data and the cost geometric correlation information to the component correlation diagram, use a dual-channel convolutional neural network to form a fusion feature map, perform data analysis on the fusion feature map, and generate a cost deviation probability distribution. The update module is used to generate a cost compensation strategy that matches the current construction status based on the cost deviation probability distribution using a reinforcement learning algorithm, and automatically transmit and update the cost compensation strategy to the BIM database to achieve dynamic adjustment of the project cost data. The step of generating a cost compensation strategy that matches the current construction status using a reinforcement learning algorithm based on the cost deviation probability distribution, and automatically transmitting and updating the cost compensation strategy to the BIM database, includes: Based on the cost deviation probability distribution, the expected benefits of different cost adjustment actions are evaluated using a reinforcement learning algorithm. Based on the expected benefits, the optimal cost adjustment action is selected to generate a cost compensation strategy. The cost compensation strategy is encapsulated in a format that meets the data structure requirements of the BIM database. The encapsulated cost compensation strategy is written into the BIM database through a data interface, so as to automatically update the engineering cost data of relevant components based on the cost compensation strategy. The process of mapping the three-dimensional deformation data and the cost geometric correlation information together onto the component correlation graph, forming a fused feature map using a dual-channel convolutional neural network, and performing data analysis on the fused feature map to generate a cost deviation probability distribution includes: The three-dimensional deformation data is mapped to the component association diagram according to the component nodes to form a deformation feature layer; The cost geometric correlation information is mapped to the component correlation graph according to the component nodes to form a cost feature layer; The deformation feature layer and the cost feature layer are input into a dual-channel convolutional neural network. Through the first convolutional channel of the dual-channel convolutional neural network, multi-scale feature extraction is performed on the deformation feature layer to obtain high-level deformation features under different receptive fields. By using the second convolutional channel of the dual-channel convolutional neural network, multi-scale feature extraction is performed on the cost feature layer to obtain high-level cost features under different receptive fields; The high-level features of deformation and the high-level features of cost are concatenated along the feature dimension through the feature fusion layer of a dual-channel convolutional neural network to generate a fused feature map. By using the fully connected layer of a dual-channel convolutional neural network, based on historical cost deviation data, the preset overall deviation range is divided into multiple cost deviation ranges. The probability values ​​of the fused feature map falling into each cost deviation interval are calculated through the Softmax activation function layer of a dual-channel convolutional neural network to generate a cost deviation probability distribution.

7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the BIM-based engineering cost management method as described in any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the BIM-based engineering cost management method as described in any one of claims 1 to 5.

Images