A Smart Lightweight Processing Method for Multi-Level Classification BIM Models

Abstract

This invention discloses a multi-level classification-based intelligent lightweighting method for BIM models. The method first separates attribute information from geometric model information in a BIM model built on Autodesk Revit, associating component IDs with their corresponding attributes. Next, it performs multi-level classification of components in the geometric model. Then, using an established optimal reduction factor classification model for component arc surfaces, it predicts the optimal reduction factor for each arc surface of the component. Based on the obtained optimal reduction factors for each arc surface, it calculates the component's reduction factor. Finally, it uses the reduction factors of different components to convert the BIM model into glTF format, achieving intelligent lightweighting of the Revit model. This invention employs different reduction factors for BIM model components of different disciplines and forms, preserving all attribute information while simultaneously improving model display quality and efficiency in BIM model lightweighting.

Description

Technical Field

[0001] This invention belongs to the technical field of Building Information Modeling (BIM), specifically relating to an intelligent lightweight processing method for multi-level classified BIM models. Background Technology

[0002] Building Information Modeling (BIM), as a three-dimensional information management technology applied throughout the entire building lifecycle, is an indispensable technological link in promoting the informatization development of the construction industry. In its application, lightweight processing of BIM models is of great significance for model storage, transmission, subsequent operation and maintenance, and especially for virtual scene display. At the same time, lightweight BIM model technology can provide excellent technical support for cloud-based management of construction processes in building engineering projects.

[0003] The research on lightweight BIM model technology aims to solve problems such as unsatisfactory terminal display caused by the complexity of BIM model component features and the large amount of model drawing data. Currently, mainstream lightweight technologies still have some problems in simplifying BIM models. For example, important information may be lost due to the deletion of redundant components and the reduction of model size; or important detailed features may be lost after the geometric model is reduced in size, resulting in phenomena such as voids and distortions on the model surface.

[0004] Converting Revit BIM models to glTF format is a commonly used lightweighting method in this field. The "surface reduction factor" refers to a parameter used in this lightweighting method to control the level of detail in the glTF 3D model. When exporting a glTF file, a parameter controls the level of detail displayed for the geometric model. Subdividing the geometric model into faces will result in different levels of detail in the view. Adjusting this parameter will change the corresponding model file size and model detail. A smaller parameter results in a smaller model file and lower model precision, while a larger parameter results in a larger model file and higher model precision. Therefore, adjusting this parameter will subdivide and reduce the number of faces in the model, thus achieving a lightweighting effect. This parameter is called the "surface reduction factor." The display problems encountered in the aforementioned lightweighting of BIM models often occur on irregularly shaped components. Since irregular curved surfaces in irregularly shaped components are mostly composed of multiple arc surfaces, optimal face reduction processing is needed for the arc surfaces of each irregularly shaped component. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides a multi-level classification-based intelligent lightweighting method for BIM models. This method employs different surface reduction coefficients for BIM model components of different disciplines and forms, ensuring that the model's display effect is considered while retaining all attribute information and achieving lightweighting.

[0006] The technical problem solved by this invention is achieved through the following technical solution: A multi-level classified intelligent lightweight processing method for BIM models is designed, characterized by comprising the following steps:

[0007] Step 1: Separate the attribute information and geometric model information of a complete BIM model built with Autodesk Revit software. Store the attribute information in a separate database and save the geometric model information as a separate Revit file. Then, associate the component IDs in the geometric model with the corresponding attribute information database.

[0008] Step 2: Based on the attribute information, perform multi-level classification of the components in the geometric model, and determine whether each component in each sub-category file contains a circular arc surface. Obtain the geometric feature parameters of each circular arc surface of the component containing a circular arc surface in each sub-category file.

[0009] Step 3: Establish a classification model for the optimal reduction coefficient of the arc surface of the architectural model components.

[0010] Step 3.1 Using the angle, radius, and height of the arc surface as geometric feature parameters, draw the corresponding BIM model in Autodesk Revit software based on the feature parameter values. Then, convert the BIM model into glTF format according to different reduction coefficients. When the included angle between two adjacent sides of the polygon is greater than or equal to 156°, the reduction coefficient at this time is the optimal reduction coefficient for that set of geometric feature parameters, and it is used as the classification label for that set of geometric feature parameters. Obtain the classification label for each set of geometric feature parameters in sequence to obtain the database. One piece of data in the database is the angle, radius, and height of the arc surface, as well as the corresponding classification label. Randomly and non-repeatingly divide the data in the database into a training set and a test set in a ratio of 8:2.

[0011] Step 3.2 Establish the optimal surface reduction coefficient classification model for arc surfaces

[0012] Based on the gradient boosting decision tree algorithm, a classification model for the optimal reduction coefficient of the arc surface is established, and hyperparameter optimization is performed using grid search and cross-validation. The geometric feature parameters X obtained in step 3.1 are used as the input variables of the gradient boosting decision tree, and the optimal reduction coefficient Y is used as the expected output to establish a gradient boosting decision tree prediction model. In the modeling process, a set of hyperparameter decision tree numbers M, a set of learning rates v, and a set of maximum leaf node numbers J for each decision tree are selected. Grid search technology is used to traverse all possible hyperparameter combinations. According to the minimum log loss criterion, the optimal parameters of the model are determined by the 5-fold cross-validation method for the data in the training set, thus obtaining the classification model for the optimal reduction coefficient of the arc surface.

[0013] Step 3.3 Use the test set to test the optimal subtraction coefficient classification model of the arc surface obtained in Step 3.2. When the accuracy is greater than 0.9, the classification model is verified to be an effective model.

[0014] Step 4: Input the geometric feature parameters of each arc surface of the component containing arc surfaces in each sub-category file in Step 2 into the classification model verified as valid in Step 3 to obtain the optimal reduction coefficient prediction value for each arc surface, and calculate the reduction coefficient for each component. The calculation method for the reduction coefficient K of a component is as follows:

[0015]

[0016] Where K is the surface reduction coefficient assigned to a component; n is the number of arc surfaces on the component containing arc surfaces;

[0017] Step 5: Based on the area reduction factor of each component calculated in Step 4, convert the BIM model built in Step 1 based on Autodesk Revit into glTF format to achieve intelligent lightweight processing of the Revit model.

[0018] Compared with existing technologies, the advantages of this invention are as follows: This invention provides a multi-level classification-based intelligent lightweighting method for BIM models. First, it performs multi-level classification on the BIM model built in Revit. Then, it uses an established classification model of optimal reduction coefficients for the arc surfaces of components to predict the optimal reduction coefficient for each arc surface of the component. Next, it calculates the reduction coefficient of the component based on the obtained optimal reduction coefficients for each arc surface. Finally, it uses the reduction coefficients of different components to convert the BIM model built based on Autodesk Revit into glTF format, achieving intelligent lightweighting of the Revit model. This invention uses different reduction coefficients for BIM model components of different disciplines and forms. While retaining all attribute information, it balances model lightweighting with improved model display quality, thus improving the efficiency and quality of BIM model lightweighting. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating an embodiment of the intelligent lightweight processing method for multi-level classified BIM models according to the present invention.

[0020] Figure 2 This is a structural diagram of seven components with different shapes in the architectural model category of the Revit BIM model in Embodiment 1 of the intelligent lightweight processing method for multi-level classification BIM models of the present invention.

[0021] Figure 3This is a schematic diagram of a near-circle regular polygon that determines the optimal reduction coefficient boundary (the included angle between two adjacent sides of the polygon is equal to 156°) in the intelligent lightweight processing method for multi-level classification of BIM models according to the present invention.

[0022] Figure 4 The accuracy of prediction results obtained by the Optimal Reduction Coefficient Classification Model (GB) for the arc surface of the intelligent lightweight processing method for multi-level classification BIM models of this invention, and the Support Vector Machine (SVM), Gaussian Process (GP), Random Forest (RF), Adaboost Iterative Algorithm (Adaboost), and Naive Bayes (NB) algorithms based on the training and test sets in Example 1.

[0023] Figure 5 This is a partial detail view of component 7 in the Revit BIM model's building model category in Example 1, shown in Revit.

[0024] Figure 6 This is a partial detail image of component 7 in the architectural model category of the Revit BIM model in Example 1, after being processed by the intelligent lightweight processing method of the multi-level classification BIM model of the present invention, and displayed on the web page. Detailed Implementation

[0025] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0026] This invention provides an intelligent lightweight processing method for multi-level classified BIM models, which includes the following steps:

[0027] Step 1: Separate the attribute information and geometric model information of a complete BIM model built with Autodesk Revit software. Store the attribute information in a separate database and save the geometric model information as a separate Revit file. Then, associate the component IDs in the geometric model with the corresponding attribute information database.

[0028] Step 2: Based on the attribute information, perform multi-level classification of the components in the geometric model, and determine whether each component in each sub-category file contains a circular arc surface. Obtain the geometric feature parameters of each circular arc surface of the component containing a circular arc surface in each sub-category file.

[0029] Step 3: Establish a classification model for the optimal reduction coefficient of the arc surface of the architectural model components.

[0030] Step 3.1 Using the angle, radius, and height of the arc surface as geometric feature parameters, draw the corresponding BIM model in Autodesk Revit software based on the feature parameter values. Then, convert the BIM model into glTF format according to different reduction coefficients. When the included angle between two adjacent sides of the polygon is greater than or equal to 156°, the reduction coefficient at this time is the optimal reduction coefficient for this set of geometric feature parameters (i.e., the criterion for the optimal reduction coefficient is that the included angle between two adjacent sides of the polygon is greater than or equal to 156°), and use it as the classification label for this set of geometric feature parameters. Sequentially obtain the classification labels for each set of geometric feature parameters to obtain the database. One piece of data in the database is the angle, radius, and height of the arc surface, as well as the corresponding classification label. Randomly and non-repeatingly divide the data in the database into a training set and a test set in a ratio of 8:2.

[0031] Step 3.2 Establish the optimal surface reduction coefficient classification model for arc surfaces

[0032] A classification model for the optimal reduction coefficient of the arc surface is established based on the gradient boosting decision tree algorithm, and hyperparameter optimization is performed using grid search and cross-validation. The geometric feature parameters X obtained in step 3.1 are used as the input variables of the gradient boosting decision tree, and the optimal reduction coefficient Y is used as the expected output to establish a gradient boosting decision tree prediction model. In the modeling process, the set of the number of hyperparameter decision trees M, the set of learning rates v, and the set of the maximum number of leaf nodes J of each decision tree are selected. All possible combinations of hyperparameters are traversed using grid search technology. According to the minimum log loss criterion, the optimal parameters of the model are determined by the 5-fold cross-validation method for the data in the training set, and the classification model for the optimal reduction coefficient of the arc surface is obtained. (The process of obtaining the classification model for the optimal reduction coefficient of the arc surface using grid search and cross-validation based on the gradient boosting decision tree algorithm is the existing technology. The implementation process can be referred to in the literature: Yan Jia, Huang Yi, Wang Xiaona. Prediction of pipeline corrosion rate based on cross-validation gradient boosting decision tree [J]. Corrosion and Protection, 2021, 42(11), 68-74.)

[0033] The formula for the logarithmic loss function (loss function) is:

[0034]

[0035] Where Y is the expected value of the optimal facet reduction coefficient, X is the geometric feature parameter, and L is the loss function. N is the sample size in the training set; M is the number of possible classes, which is obtained statistically based on the value of the optimal facet reduction coefficient Y in the training set. When using the 5-fold cross-validation method, the training set is divided into 5 datasets, then M is obtained statistically based on the value of the optimal facet reduction coefficient Y in the corresponding dataset; y ij It is a binary index indicating whether category j is a geometric feature parameter x. iThe true category is determined by assigning 1 to true and 0 to false; p ij Predict geometric feature parameters x for the model i The probability of belonging to category j.

[0036] Step 3.3 Use the test set to test the optimal subtraction coefficient classification model of the arc surface obtained in Step 3.2. When the accuracy (i.e., the proportion of accurately predicted samples to the total number of predicted samples) is greater than 0.9, the classification model is verified as an effective model.

[0037] Step 4: Input the geometric feature parameters of each arc surface of the component containing arc surfaces in each sub-category file in Step 2 into the classification model verified as valid in Step 3 to obtain the optimal reduction coefficient prediction value for each arc surface, and calculate the reduction coefficient for each component. The calculation method for the reduction coefficient K of a component is as follows:

[0038]

[0039] Where K is the reduction factor assigned to a component; n is the number of arc surfaces on the component containing arc surfaces.

[0040] Step 5: Based on the area reduction factor of each component calculated in Step 4, convert the BIM model built in Step 1 based on Autodesk Revit into glTF format to achieve intelligent lightweight processing of the Revit model.

[0041] Example 1

[0042] This embodiment provides a method for intelligent lightweight processing of multi-level classified BIM models, which includes the following steps:

[0043] Step 1: Separate the attribute information and geometric model information of the hotel BIM model built using Autodesk Revit. Store the attribute information in a separate database, and save the geometric model information as a separate Revit file. Link the geometric model's component IDs to the corresponding attribute information database. Using the secondary development interface provided by Revit, attribute information can be quickly extracted from the BIM model and then stored in a separate relational database. The geometric model information is then saved as a separate Revit file, and the component IDs in the geometric model are linked to the corresponding attribute information data.

[0044] Step Two: Based on attribute information, perform multi-level classification of components in the geometric model. The geometric model obtained after separation in Step One is first classified into six files according to their respective specialties: architecture, structure, HVAC, electrical, power, and water supply / drainage, and named accordingly. To illustrate the method of this invention, the architectural model is selected, and then seven components with different shapes are selected and named "Floor Slab.rvt". Figure 2 As shown, components 1, 2, and 3 are components of different sizes without curved surfaces; components 4 and 5 are components of different sizes with one curved surface; and components 6 and 7 are components of different sizes with different numbers of curved surfaces. Then, it is determined whether each component's geometric model file contains curved surfaces, and based on the determination result, the components are divided into three categories: no curved surfaces (components 1, 2, and 3), one curved surface (components 4 and 5), and multiple curved surfaces (components 6 and 7). Furthermore, the curved surfaces in components containing multiple curved surfaces are numbered, and the geometric feature parameters of all curved surfaces are obtained.

[0045] A program using conditional statements and loops is written to perform multi-level classification of components in a geometric model based on corresponding attribute information, including model specialty and component category. Next, it determines whether each component in each specialty's geometric model file contains curved surfaces, and further classifies the components based on the determination result. For components containing curved surfaces, their curved surfaces and corresponding geometric feature parameters are further obtained.

[0046] When a BIM model created using Revit is delivered, it is categorized according to different disciplines based on design rules. Using ElementFilters to traverse the components and their corresponding component IDs in the model, the component's category information can be obtained. Whether a component contains a curved surface is determined by checking if its attribute information includes "CylindricalFace". Taking an extruded body in a Revit BIM model as an example, the geometric feature parameters of a curved surface, including radius, angle, and height, are obtained using the following method: The method for obtaining the geometric feature parameters of a curved surface is as follows:

[0047] When obtaining the value of the radius of the arc surface:

[0048] First, retrieve the Geometry property from the Element attribute.<Geometry.Element> Value, and then get <geometryelement>Select Solid to continue obtaining Faces. <facearray>The value is then used to retrieve the EdgeLoops value. <edgearrayarray>Select the Radius property of Arc, and then convert the imperial units to metric units.

[0049] When obtaining the angle value:

[0050] The angle value of the arc surface cannot be obtained directly, but can be obtained indirectly through calculation. The first few steps are the same as when obtaining the radius of the arc surface. <edgearrayarray>Next, select the values ​​of the Radius and ApproximateLength attributes of Arc to convert imperial units to metric units. Then, according to the angle arc length formula L=nπr / 180, we can get n=180L / πr, where L=ApproximateLength and r=Radius.

[0051] When obtaining the thickness value:

[0052] Retrieving Parameters from Element Properties <parameterset>You can find the "unconnected height" or "thickness" in the settings, and then convert the imperial units to metric units.

[0053] Step 3: Establish a classification model for the optimal reduction coefficient of the arc surface of the building model components.

[0054] In the Revit model, the face reduction factor ranges from 0 to 15, making the selection of the optimal face reduction factor a classification problem.

[0055] Step 3.1 Using the angle, radius, and height of the arc surface as geometric feature parameters, draw the corresponding BIM model in Autodesk Revit software based on the feature parameter values. Then, convert the BIM model into glTF format according to different reduction coefficients. When the included angle between two adjacent sides of the polygon is greater than or equal to 156°, the reduction coefficient at this time is the optimal reduction coefficient for that set of geometric feature parameters, and use it as the classification label for that set of geometric feature parameters. Sequentially obtain the classification labels for each set of geometric feature parameters to obtain a database. One piece of data in the database is the angle, radius, and height of the arc surface, as well as the corresponding classification label. Randomly and non-repeatingly divide the data in the database into a training set and a test set in a ratio of 8:2.

[0056] The angle, radius, and height of the arc surface were used as geometric feature parameters. The radius values ​​were 100, 200, 300, 400, 500, 600, 700, 800, 900, 1000, 1100, 1200, 1300, 1400, 1500, 1600, 1700, 1800, 1900, 2000, 3000, 4000, and 5000, in millimeters; the angle values ​​were 360, 330, 300, 270, 240, 210, 180, 150, 120, 90, 60, 45, and 30, in degrees; and the height values ​​were 100, 500, 1000, 2000, and 5000, in millimeters. A total of 1495 samples were established.

[0057] To determine the optimal facet reduction factor for these samples and use it as a classification label, a corresponding BIM model was drawn in Revit based on the feature parameter values. Then, the model was converted to glTF format using a lightweight method according to different facet reduction factors, and viewed on a web page using the Three.js framework. As the facet reduction factor increases, the curved surface gradually becomes polygonal. It is generally considered that when the included angle between adjacent sides of the polygon is greater than or equal to 156°, it does not affect the visual effect of the curved surface. Figure 3 As shown, this is used as the critical value to determine the optimal surface reduction coefficient. The optimal surface reduction coefficient of a set of geometric feature parameters is used as its classification label; the classification label of each set of geometric feature parameters is obtained in turn to obtain the database; a data in the database is the angle, radius and height of the arc surface, and the corresponding classification label; the data in the database are randomly and non-repeatingly divided into training set and test set in a ratio of 8:2; the constructed database is saved as a CSV file.

[0058] Step 3.2 Establish the optimal surface reduction coefficient classification model for arc surfaces

[0059] A classification model for the optimal reduction coefficient of the arc surface is established based on the gradient boosting decision tree algorithm, and hyperparameter optimization is performed using grid search and cross-validation. The geometric feature parameters X obtained in step 3.1 are used as the input variables of the gradient boosting decision tree, and the optimal reduction coefficient Y is used as the expected output to establish a gradient boosting decision tree prediction model. In the modeling process, the set of the number of hyperparameter decision trees M, the set of learning rates v, and the set of the maximum number of leaf nodes J of each decision tree are selected. All possible combinations of hyperparameters are traversed using grid search technology. According to the minimum log loss criterion, the optimal parameters of the model are determined by the 5-fold cross-validation method for the data in the training set, and the classification model for the optimal reduction coefficient of the arc surface is obtained. (The process of obtaining the classification model for the optimal reduction coefficient of the arc surface using grid search and cross-validation based on the gradient boosting decision tree algorithm is the existing technology. The implementation process can be referred to in the literature: Yan Jia, Huang Yi, Wang Xiaona. Prediction of pipeline corrosion rate based on cross-validation gradient boosting decision tree [J]. Corrosion and Protection, 2021, 42(11), 68-74.)

[0060] The formula for the logarithmic loss function (loss function) is:

[0061]

[0062] Where Y is the expected value of the optimal facet reduction coefficient, X is the geometric feature parameter, and L is the loss function. N is the sample size in the training set; M is the number of possible classes, which is obtained statistically based on the value of the optimal facet reduction coefficient Y in the training set. When using the 5-fold cross-validation method, the training set is divided into 5 datasets, then M is obtained statistically based on the value of the optimal facet reduction coefficient Y in the corresponding dataset; y ij It is a binary index indicating whether category j is a geometric feature parameter x. i The true category is determined by assigning 1 to true and 0 to false; p ij Predict geometric feature parameters x for the model i The probability of belonging to category j.

[0063] Step 3.3 Use the test set to test the optimal subtraction coefficient classification model of the arc surface obtained in Step 3.2. When the accuracy (i.e., the proportion of accurately predicted samples to the total number of predicted samples) is greater than 0.9, the classification model is verified as an effective model.

[0064] To evaluate the performance of the arc surface optimal reduction coefficient classification model designed in this invention, other machine learning algorithms were introduced and tested on the training and test sets. These algorithms included Support Vector Machine, Gaussian Process, Random Forest, AdaBoost iterative algorithm, and Naive Bayes. Accuracy was used as the evaluation metric for model prediction performance, and the results are as follows: Figure 4 As shown. Figure 4 In this diagram, SVM (Support Vector Machines) refers to the Support Vector Machine algorithm, GP (Gaussian Process) refers to the Gaussian Process algorithm, RF (Random Forest) refers to the Random Forest algorithm, Adaboost refers to the Adaboost iterative algorithm, GB (Gradient Boosting) refers to the optimal reduction coefficient classification model for circular arc surfaces designed in this invention (based on the gradient boosting algorithm), and NB (Naive Bayes) refers to the Naive Bayes algorithm. The y-axis AccuracyScore in the diagram represents the accuracy of the model trained by each algorithm. Figure 4 As can be seen, the classification model for the optimal reduction coefficient of the arc surface designed in this invention has the highest prediction accuracy.

[0065] Step 4: Input the geometric feature parameters of each arc surface of the component containing arc surfaces in each sub-category file in Step 2 into the classification model verified as valid in Step 3 to obtain the optimal reduction coefficient prediction value for each arc surface, and calculate the reduction coefficient for each component. The calculation method for the reduction coefficient K of a component is as follows:

[0066]

[0067] Where K is the reduction factor assigned to a component; n is the number of arc surfaces on the component containing arc surfaces.

[0068] According to the prediction results of the optimal reduction coefficient classification model, when the optimal reduction coefficients for the six arc surfaces of component 6 are 5, 5, 5, 4, 6, and 5 respectively, the optimal reduction coefficient for this component is 6. The optimal reduction coefficients for the eight arc surfaces of component 7 are 7, 7, 5, 6, 6, 6, 7, and 5 respectively, and the optimal reduction coefficient for this component is 7.

[0069] Among them, components 1, 2, and 3, which do not have arc surfaces, are assigned a reduction factor of 0; components 4, 5, and 6 have reduction factors of 7, 5, and 6 respectively according to the calculation results.

[0070] Step 5: Based on the area reduction factor of each component calculated in Step 4, convert the BIM model built in Step 1 based on Autodesk Revit into glTF format to achieve intelligent lightweight processing of the Revit model.

[0071] When all components in the entire model are assigned a specific reduction factor, the floor slab .rvt format file will be exported as floor slab .gltf and floor slab .bin files, which can be viewed on a web browser. Figure 5 This is a partial detail view of component 7 in the RevitBIM model's building model category in this embodiment. Figure 6 In this embodiment, component 7 in the architectural model category of the Revit BIM model is displayed as a partial detail image on the webpage after being processed by the intelligent lightweight processing method for multi-level classification BIM models according to the present invention. Figure 6 As can be seen, there are no issues such as distortion, broken surfaces, or voids. The surface and boundary features of the components can be observed. For the arc surface, if it appears as a polygon when magnified, it approximates an arc when the viewing angle is pulled back. The visual effect retains the feature-rich area. Moreover, the original model is 5478kb, and after being lightweighted by this method, it is only 208kb, which can be loaded and displayed smoothly on the web.

[0072] By employing secondary development techniques and leveraging the official Revit API, a plugin is developed to extend Revit's format conversion capabilities. This implements the IExternalCommand secondary development interface, extracting component information, identifying curved surfaces, acquiring information, and providing the aforementioned multi-level classification functionality. Utilizing the IExportContext interface for format conversion essentially maps Revit component information to the glTF framework, enabling data export and conversion within Revit files.

[0073] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0074] Any aspects not covered in this invention are applicable to existing technologies.< / parameterset> < / edgearrayarray> < / edgearrayarray> < / facearray> < / geometryelement>

Claims

1. A method for intelligent lightweight processing of a multi-level classified BIM model, characterized in that, The method comprises the following steps: Step 1: Separate the attribute information and the geometric model information of a complete BIM model established based on Autodesk Revit software, store the attribute information in an independent database, store the geometric model information as an independent Revit file, and associate the geometric model information with the corresponding attribute information database through the ID of the component in the geometric model; Step 2: According to the attribute information, classify the components in the geometric model in multiple levels, and determine whether each component in each subcategory file contains an arc surface, and obtain the geometric feature parameters of each arc surface of each component containing an arc surface in each subcategory file; Step 3.1: Take the angle, radius and height of the arc surface as the geometric feature parameters, draw the corresponding BIM model in the Autodesk Revit software according to the feature parameter values, then convert the BIM model into a glTF format according to different reduction coefficients, when the included angle between two adjacent polygons is greater than or equal to 156°, the reduction coefficient at this time is the optimal reduction coefficient of the group of geometric feature parameters, which is taken as the classification label of the group of geometric feature parameters; obtain the classification label of each group of geometric feature parameters in turn to obtain a database; one piece of data in the database is the angle, radius and height of the arc surface, and the corresponding classification label; divide the data in the database into a training set and a test set in a random and non-repetitive manner according to the quantity ratio of 8:2; Step 3.2: Establish the optimal reduction coefficient classification model of the arc surface On the basis of the gradient boosting decision tree algorithm, the optimal reduction coefficient classification model of the arc surface is established, and the grid search and cross-validation method is used for hyperparameter optimization; the geometric feature parameters X in the training set obtained in step 3.1 are used as the input variables of the gradient boosting decision tree, and the optimal reduction coefficient Y is used as the expected output to establish a gradient boosting decision tree prediction model; when modeling, the set of the number of decision trees M, the set of the learning rate v and the set of the maximum number of leaf nodes J of each decision tree are selected, all possible combinations of hyperparameters are traversed by using the grid search technology, and the optimal parameters of the model are determined according to the minimum log loss criterion by using the 5-fold cross-validation method for the data in the training set, so as to obtain the optimal reduction coefficient classification model of the arc surface; Step 3.3: Test the optimal reduction coefficient classification model of the arc surface obtained in step 3.2 by using the test set, when the accuracy is greater than 0.9, it is verified that the classification model is an effective model; Step 4: Input the geometric feature parameters of each arc surface of each component containing an arc surface in each subcategory file in step 2 into the effective classification model in step 3, obtain the optimal reduction coefficient prediction value of each arc surface, and calculate the reduction coefficient of each component; the calculation method of the reduction coefficient K of a component is as follows: Wherein, K is the reduction coefficient of a component; n is the number of arc surfaces on the component containing an arc surface. ​ Step five: According to the reduction factor of each component calculated in step four, the BIM model based on Autodesk Revit in step one is converted into glTF format, realizing the intelligent lightweight processing of Revit model.

2. The multi-level classified BIM model intelligent lightweight processing method according to claim 1, characterized in that, The formula of log loss function is: Wherein, Y is the expected value of the optimal reduction surface coefficient, X is the geometric feature parameter, L is the loss function, N is the sample size in the training set, M is the possible number of categories, is obtained according to the statistics of the value of the optimal reduction surface coefficient Y in the training set, when the 5-fold cross-validation method is used, the training set is divided into 5 data sets, then M is obtained according to the statistics of the value of the optimal reduction surface coefficient Y in the corresponding data set; y ij is a binary index indicating whether the category j is the true category of the geometric feature parameter x i , and is 1 when judged as true and 0 when judged as false; p ij is the probability of the model predicting that the geometric feature parameter x i belongs to the category j.

3. The multi-level classified BIM model intelligent lightweight processing method according to claim 1, characterized in that, The geometric characteristic parameters of a circular arc surface of a component are obtained as follows: When the value of the radius of the arc surface is obtained: First, the <Geometry.Element> value of the Geometry in the Element property is acquired, and then the <Geometry.Element> value of the Geometry in the Geometry property is acquired <geometryelement>, select Solid from the list, and continue to get Faces of the Solid <facearray>values, and continue to acquire the EdgeLoops therein <edgearrayarray>The value of the attribute Radius of Arc is selected, and the units are converted from inch to meter;< / edgearrayarray> < / facearray> < / geometryelement> When the value of the angle is obtained: The angle value of the circular arc surface cannot be directly obtained and can be indirectly obtained by calculation. The operations in the previous steps are the same as those when obtaining the value of the radius of the arc surface. In the process of obtaining the EdgeLoops <edgearrayarray>After that, the value of the attribute Radius and the value of the attribute ApproximateLength of Arc are selected, the units are converted from inch to meter, and according to the formula of the arc length according to the angle L=nπr / 180, n=180L / πr can be obtained, where L=ApproximateLength and r=Radius;< / edgearrayarray> When the value of the thickness is obtained: Get the Parameters in the Element property <parameterset>The value of "unconnected height" or "thickness" in the above table can be selected, and the units are converted from inch to meter.< / parameterset>

Images