A BIM-based digital twin model creation method

Abstract

The application discloses a BIM-based digital twin model creation method, comprising: S1 semantic fingerprint evaluation and completion, calculating a semantic fingerprint vector from three dimensions of geometric integrity, attribute integrity and relationship integrity to identify and complete an unready component; S2 intention graph clipping and enhancement, querying an intention graph according to a target application scenario to clip a minimized submodel and performing directional semantic enhancement; S3 semantic geometry decoupling construction, extracting a building topology skeleton to construct a semantic topology layer and a geometric performance layer respectively, and outputting a decoupled multi-resolution digital twin model; and S4 function semantic inference injection, inferring implicit function information based on a function space semantic graph and a building function knowledge base and injecting the semantic topology layer. The application improves the data quality, scene adaptability and semantic depth of the digital twin model.

Description

Technical Field

[0001] This invention relates to the field of building information modeling and digital twin technology, and in particular to a method for creating a digital twin model based on BIM. Background Technology

[0002] Building Information Modeling (BIM), as the core carrier of building lifecycle data, provides the basic data source for the construction of digital twins. Digital twin technology requires mapping physical buildings into virtual models with semantic awareness and multi-resolution expression capabilities to support various application scenarios such as energy consumption analysis, fire safety assessment, and equipment operation and maintenance.

[0003] Existing methods for converting BIM models to digital twin models mainly face the following technical challenges: First, existing methods typically import the entire BIM model into the digital twin platform, lacking a quantitative assessment mechanism for the readiness of component-level data. When the BIM model has quality issues such as missing geometry, incomplete attributes, or broken topological relationships, direct conversion will make it difficult to guarantee the semantic consistency of the digital twin model, thus affecting the accuracy of subsequent applications.

[0004] Second, different digital twin application scenarios have significantly different requirements for BIM data. For example, energy consumption analysis scenarios focus on the thermal performance parameters of the building envelope, while fire safety scenarios emphasize evacuation distances and fire compartment information. Existing methods mostly adopt a uniform conversion strategy, which cannot reduce the model size by trimming redundant data according to specific scenarios, nor can it supplement the semantic information not explicitly included in the BIM model to meet the needs of the scenario. This results in the generated digital twin model having both data redundancy and semantic deficiencies.

[0005] Third, traditional digital twin models often couple and store geometric data with semantic data. When switching between different levels of detail, the complete model needs to be reloaded, resulting in low rendering efficiency and difficulty in adapting to the differentiated needs of multi-terminal access.

[0006] Fourth, BIM models typically only contain information about explicitly modeled components, lacking inference mechanisms for implicit functional information such as spatial functions, personnel capacity, and equipment design load, which limits the depth of application of digital twin models in intelligent operation and maintenance decision-making.

[0007] Therefore, a systematic approach is needed that can generate high-quality, scenario-adaptive, and semantically rich digital twin models, encompassing data quality assessment, scenario-driven pruning, decoupling of multi-resolution representations, and functional semantic inference. Summary of the Invention

[0008] The purpose of this invention is to provide a method for creating digital twin models based on BIM, in order to solve the technical problems in the conversion of BIM models to digital twin models in the prior art, such as uncontrollable data quality, insufficient scene adaptability, geometric and semantic coupling, and lack of implicit functional information.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A method for creating a BIM-based digital twin model includes the following steps: S1: Semantic fingerprint assessment and completion. For each component in the BIM model, an assessment score is calculated from three dimensions: geometric completeness, attribute completeness, and relational completeness, and these scores are combined into a semantic fingerprint vector. The semantic fingerprint vector is then compared with a pre-constructed readiness standard fingerprint. Components with a similarity score below the readiness threshold are identified as incomplete components, and their missing information is completed based on a component type knowledge base. The semantic fingerprint vector is a comprehensive feature vector obtained by quantifying the completeness of a component in the three dimensions of geometric representation, attribute definition, and relational instance. S2: Intent Graph Pruning and Enhancement. Based on the target application scenario, a pre-built intent graph is queried to extract the corresponding data requirement description. According to the data requirement description, a minimized sub-model containing only the required components and their attributes is pruned from the BIM model processed in step S1. Targeted semantic enhancement is then performed on the minimized sub-model. The intent graph is a graph structure that organizes the digital twin application scenario into intent nodes, each associated with the data requirement description. Targeted semantic enhancement involves supplementing the minimized sub-model with attribute information required by the target application scenario but not included in the BIM model. The BIM model in step S2 is the model after quality completion in step S1, ensuring that the pruning operation is performed based on a quality-controllable data foundation. S3: Semantic-geometric decoupling construction. An architectural topological skeleton is extracted from the minimized sub-model. A semantic topological layer is constructed using the architectural topological skeleton and component attribute information. The geometric data of each component is independently encoded into a geometric representation layer according to multi-level detail. A multi-resolution digital twin model, decoupled between the semantic topological layer and the geometric representation layer, is output. The architectural topological skeleton is a graph structure with spaces and components as nodes and architectural topological relationships as edges. Step S3 uses the minimized sub-model output in step S2 as input, ensuring that the extraction range of the topological structure has converged according to the scene requirements. S4: Functional semantic inference injection. A functional space semantic graph is extracted from the building topology skeleton. Based on the correspondence between space usage and functional requirements in the building function knowledge base, the implicit functional information not explicitly modeled in the BIM model is inferred from the space usage attributes of each node in the functional space semantic graph. The inference results are then injected into the semantic topology layer. The functional space semantic graph is a graph structure with rooms or spaces as nodes, accessibility relationships as edges, and each node labeled with a space usage attribute. The inference results from step S4 are injected into the semantic topology layer constructed in step S3, allowing the digital twin model to obtain supplementary implicit functional information while maintaining a geometrically and semantically decoupled structure.

[0010] Furthermore, the geometric completeness is determined based on the shape descriptor matching score between the component's geometric expression and the corresponding level of detail standard template; the attribute completeness is determined based on the coverage of each required attribute in the standard attribute set corresponding to the component type; and the relational completeness is determined based on the instance completeness of the component in the IFC space containment relationship, component connection relationship, and system dependency relationship.

[0011] Furthermore, the step of supplementing the missing information of the unready component based on the component type knowledge base specifically includes: retrieving the attribute value distribution of historical components of the same type as the unready component from the component type knowledge base, using the attribute value with the highest probability of occurrence in the attribute value distribution as the recommended supplement value, and writing it into the BIM model after user confirmation.

[0012] Furthermore, the data requirement description includes the set of component types, the set of attribute fields, and the level of detail required by the target application scenario; when performing pruning in step S2, components whose component types belong to the set of component types and whose similarity calculated in step S1 is not lower than the readiness threshold are retained.

[0013] Furthermore, the targeted semantic enhancement applies different enhancement rules according to the type of the target application scenario. The enhancement rules include: supplementing the thermal parameters of the building envelope for energy consumption analysis scenarios, marking evacuation distances and fire compartment information for fire safety scenarios, and supplementing the range of equipment operating parameters for equipment operation and maintenance scenarios.

[0014] Furthermore, the building topology includes spatial connectivity represented by doors and windows, system topology represented by pipeline connections, and functional hierarchy represented by floor-area-room containment levels.

[0015] Furthermore, the multi-level detail hierarchy includes an envelope level, a simplified block level, and a fine model level; the loading level of each component is determined by weighting a view distance factor, a semantic importance factor, and a task association factor, wherein the task association factor is determined according to the detail hierarchy level specified for each component type based on the target application scenario described in the data requirement description.

[0016] Furthermore, the inference of the implicit functional information includes: inferring the personnel capacity and fire evacuation requirements based on the space use and area; inferring the lighting and air conditioning design load based on the space use; inferring the ventilation path and noise propagation chain based on the accessibility relationship; and inferring the sound insulation or heat insulation requirement level of the partition components based on the differences in the uses of adjacent spaces.

[0017] Furthermore, the inference result includes a confidence score and an inference path; inference results with a confidence score not lower than the confirmation threshold are directly injected into the semantic topology layer, while inference results with a confidence score lower than the confirmation threshold are marked as pending confirmation and submitted to the user for review, and then injected into the semantic topology layer after confirmation.

[0018] Furthermore, when multiple different target application scenarios are specified for the same BIM model, each target application scenario shares the same BIM model processed by step S1, and steps S2 to S4 are executed independently for each target application scenario to generate a corresponding digital twin model.

[0019] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: (1) By comparing the similarity between the semantic fingerprint vector and the readiness standard fingerprint, the quality of BIM model data can be quantitatively evaluated and automatically completed at the component level. It is difficult to quantify the geometric completeness, attribute completeness and relational completeness in a coordinated manner by using fixed thresholds or manual inspection. The semantic fingerprint vector method combines the evaluation scores of the three dimensions into a unified feature vector and calculates the similarity through weighted Euclidean distance. It can accurately locate the dimension where the quality defect is located and give a targeted completion strategy. Combined with the historical attribute value distribution of the same type of component in the component type knowledge base, it performs probabilistic recommendation completion, reduces the risk of low-quality data spreading to the digital twin model, and thus provides a quality-controllable data foundation for subsequent scene tailoring. (2) By introducing the scene-aware pruning mechanism driven by intent graph and combining it with the targeted semantic enhancement strategy, the generated digital twin model can simultaneously meet the requirements of data minimization and semantic completeness. The unified transformation strategy cannot organize data differently according to scene requirements and lacks a formal description mechanism for scene requirements. Intent graph encodes the set of component types, attribute fields and detail levels required by each application scene into a structured data requirement description, establishing an explicit mapping relationship between scene requirements and data structure, making the pruning process interpretable and reproducible. At the same time, targeted semantic enhancement supplements scene-specific attributes according to the scene type using differentiated enhancement rules, eliminating the problem of data redundancy and semantic loss. (3) By decoupling the semantic topology layer from the geometric representation layer, it supports independent loading and switching of multiple levels of detail. The method of integrated storage of geometry and semantics requires reloading the complete model when switching levels of detail, resulting in storage redundancy and response delay. The decoupled architecture establishes a mapping relationship between the two layers through the unique identifier of the component, so that semantic updates and geometric rendering can be executed in parallel. Moreover, the geometric representation layer can independently select the loading level based on the comprehensive score of the viewing distance factor, semantic importance factor and task association factor, which is conducive to improving the system response efficiency in multi-terminal collaborative access scenarios. (4) By using a collaborative inference mechanism based on the functional space semantic graph and the building function knowledge base, the implicit functional information that is not explicitly modeled in the BIM model is automatically supplemented. The method of building a digital twin based solely on explicit data lacks the ability to infer implicit functional information. However, this invention extracts the functional space semantic graph with spaces as nodes and accessibility relationships as edges from the building topology skeleton, and performs structured inference by combining the correspondence between space use and functional requirements in the building function knowledge base. This transforms space use and topological relationships into operable engineering decision parameters such as personnel capacity, equipment load, and sound and heat insulation requirements. Furthermore, the inference results are injected in a hierarchical manner through confidence assessment, thereby expanding the semantic depth of the digital twin model. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating a method for creating a BIM-based digital twin model, as provided in an embodiment of the present invention.

[0021] Figure 2 This diagram illustrates the data interaction relationships between the steps of a BIM-based digital twin model creation method provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0023] Reference manual attached Figure 1 The diagram illustrates a flowchart of a BIM-based digital twin model creation method provided by an embodiment of the present invention.

[0024] Reference manual attached Figure 2 This diagram illustrates the data interaction relationships between the steps provided in an embodiment of the present invention. Figure 2 As shown, the data processing flow of this embodiment of the invention takes a BIM model as the initial input and proceeds sequentially through steps S1 to S4. In step S1, the BIM model interacts with the component type knowledge base to perform semantic fingerprint evaluation on each component and complete the missing information of unready components, outputting the completed BIM model. In step S2, the completed BIM model interacts with the intent graph to extract data requirement descriptions based on the target application scenario, trims out a minimized sub-model, performs targeted semantic enhancement, and outputs the enhanced minimized sub-model. In step S3, the building topology skeleton is extracted from the minimized sub-model, and a semantic topology layer and a geometric representation layer are constructed respectively. The two layers are linked by unique component identifiers but their data organization is independent. In step S4, a functional space semantic graph is extracted from the building topology skeleton, interacts with the building function knowledge base to infer implicit functional information, and the inference results are injected into the semantic topology layer constructed in step S3. The above data interaction relationship indicates that the component type knowledge base, intent map, and building function knowledge base intervene in the data processing flow as external knowledge sources in steps S1, S2, and S4, respectively, and the steps achieve unidirectional transmission through explicit intermediate data objects.

[0025] This invention provides a method for creating a BIM-based digital twin model, which may include the following steps: S1: Semantic Fingerprint Assessment and Completion: For each component in the BIM model, an assessment score is calculated from three dimensions: geometric completeness, attribute completeness, and relational completeness. This score is then combined into a semantic fingerprint vector. The semantic fingerprint vector is compared with a pre-constructed readiness standard fingerprint. Components with a similarity score below the readiness threshold are identified as incomplete components, and their missing information is completed based on a component type knowledge base. The semantic fingerprint vector is a comprehensive feature vector obtained by quantifying the completeness of a component in the three dimensions of geometric representation, attribute definition, and relational instance.

[0026] Specifically, suppose the BIM model contains a set of components. ,in, Indicates the first Each component Indicates component index and , Indicates the total number of components and For any component Calculate the geometric completeness score respectively. Attribute completeness score Relationship integrity score The calculation of the three scores is independent of each other.

[0027] It should be noted that the component type knowledge base is built based on the component type system defined in the IFC standard. It stores the standard geometric templates, required attribute sets, relationship specifications, and attribute value statistics of the same type of components in historical projects for each component type.

[0028] Furthermore, the geometric completeness score is determined based on the shape descriptor matching score between the component's geometric representation and the corresponding level-of-detail standard template. For the component... Extracting the 3D shape descriptor vector from the geometric representation yields... The descriptor vector of the standard template corresponding to the type of this component is Geometric completeness score The calculation formula is shown in formula (1): in, Representing components Geometric completeness score and , Representing components geometric descriptor vector The Each component is dimensionless and , Represents the corresponding standard template descriptor vector The Each component is dimensionless and , Indicates the index of the descriptor component and , Indicates the dimension of the descriptor and , Represents the geometric deviation tolerance parameter and is dimensionless. , This represents an exponential function with the natural constant as its base. To represent the square root operation, This indicates that a summation operation is performed on each item within the specified index range.

[0029] It should be noted that formula (1) adopts a piecewise definition: when the component The geometric expression is completely missing, that is When it is a zero vector, The division operation in formula (1) is not performed; when the component Having an effective geometric expression When the vector is non-zero, calculate according to formula (1). ,at this time A non-zero vector such that in the denominator Standard template descriptor vector To make a non-zero vector predefined from the component type knowledge base such that The fact that the denominator is always greater than zero ensures that the division is valid. In formula (1) and All values ​​are non-negative because each component of the 3D shape descriptor represents the frequency or amplitude of the geometric features. The first part of Formula (1) is the cosine similarity, which measures the consistency of the component geometric descriptor with the standard template in direction; the second part is the Gaussian decay factor, which exponentially penalizes the Euclidean distance between descriptors. The product structure of the two makes the score reflect both the overall trend consistency of the geometric shape and a sensitive response to local shape defects or distortions. Compared with the method of using single cosine similarity, which can only measure directional consistency and is not sensitive to amplitude deviation, Formula (1) compensates for the Euclidean distance through the Gaussian decay factor, and can more comprehensively characterize the degree of geometric integrity.

[0030] Optionally, the three-dimensional shape descriptor uses a D2 descriptor, which is a normalized frequency distribution histogram of the Euclidean distance between two random points. Since each component value is a frequency value, it satisfies non-negativity. It is understood that those skilled in the art can choose other shape descriptors according to actual circumstances, and this embodiment of the invention does not specifically limit this choice.

[0031] Optionally, The value is determined based on the statistical standard deviation of geometric descriptors for similar components, and is set to 0.2. This value ensures that when the normalized Euclidean distance between the component descriptor and the standard template descriptor reaches 0.2, the Gaussian attenuation factor drops to approximately 0.61, effectively penalizing components with deviations exceeding one standard deviation. It is understood that those skilled in the art can adjust the value based on the specific circumstances. The value of is adjusted, but the embodiments of the present invention do not impose specific limitations on this.

[0032] Furthermore, the attribute completeness score is determined based on the coverage of each required attribute in the standard attribute set corresponding to the component type. Let the component... The standard property set defined by the type contains There are 1 required attributes, and the importance weight of each required attribute is: The existence indicator function for the corresponding attribute is The attribute value validity factor is Attribute completeness score The calculation formula is shown in formula (2): in, Representing components Attribute completeness score and , Representing components The number of required attributes in the standard attribute set defined by the type and , Indicates the required attribute index and , Indicates the first The importance weights of the required attributes are dimensionless and Furthermore, the degree of influence of this attribute on digital twin applications is predefined in the component type knowledge base. Indicates the first The function indicates the existence of a required attribute, taking the value 1 if the attribute exists in the BIM model and 0 if it is missing. Indicates the first The validity factor for each required attribute is 1 if the attribute value passes format validation and is within a reasonable range, and 0 otherwise. Because and The sum of the denominators is always greater than zero, which ensures that the division is valid.

[0033] It should be noted that the value validity factor is introduced in formula (2). This allows the scoring to consider not only the existence of attributes but also the actual usability of attribute values; it introduces importance weights. Distinguish the differences in contribution of core attributes and auxiliary attributes to the overall completeness. Compared with the simple coverage method, which divides the number of existing attributes by the total number of attributes, the simple coverage method treats all attributes equally and does not distinguish the validity of attribute values. Formula (2) enables the evaluation results to accurately reflect the actual support of attribute information for subsequent digital twin applications through the joint evaluation of weighted coverage and value validity.

[0034] Optionally, importance weight The component type knowledge base predefines the weights based on attribute categories, with core structural attributes having a weight of 1.0, functional performance attributes having a weight of 0.8, and auxiliary descriptive attributes having a weight of 0.5. It is understood that those skilled in the art can adjust these weight values ​​according to actual circumstances, and this embodiment of the invention does not impose specific limitations in this regard.

[0035] Furthermore, the relational integrity score is determined based on the instance integrity of components in the IFC space, component connection relationships, and system dependency relationships. Let the component... The expected number of instances of the three types of relations involved are respectively , and The actual number of valid instances are respectively , and Relationship integrity score The calculation formula is shown in formula (3): in, Representing components Relationship completeness score and , Representing components The expected number of instances in spatial containment relationships and Furthermore, the relationship specifications for the corresponding types are determined by the component type knowledge base. Representing components The expected number of instances in the component connection relationship and , Representing components The expected number of instances in the system hierarchy and , This represents the number of actual and valid instances in the spatial containment relation. , This represents the number of actual and valid instances in the component connection relationship. , This represents the number of actual and valid instances in the system's dependency relationships. , The weight coefficients representing spatial inclusion relationships are dimensionless and , The weighting coefficients representing the connection relationships between components are dimensionless and , The weight coefficients representing the system's subordinate relationships are dimensionless and And satisfy , This indicates the operation of taking the smaller of the two inputs. Because... , , A denominator that is always greater than zero ensures that the division is valid; The operation truncates the ratios of various relationships to 1 to avoid redundant relationship instances causing scores to exceed a reasonable range.

[0036] Optionally, The value is 0.4. The value is 0.35. The value is set to 0.25. This value is determined based on engineering experience, where spatial inclusion relationships are considered the most critical for the fundamental support of the building's topology, followed by the influence of component connection relationships on the physical system's expression, while the weight of system subordination relationships is relatively low but not negligible. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to actual circumstances, and this embodiment of the invention does not impose specific limitations in this regard.

[0037] Furthermore, the evaluation scores from the three dimensions are combined into a semantic fingerprint vector and its similarity is calculated with the readiness standard fingerprint. The geometric completeness score is then... Attribute completeness score Relationship integrity score Combined into components semantic fingerprint vector Let the pre-constructed readiness standard fingerprint be... Semantic fingerprint similarity The calculation formula is shown in formula (4): in, Representing components The similarity between the semantic fingerprint vector and the readiness standard fingerprint and , Representing components The geometric completeness score is calculated using formula (1). Representing components The attribute completeness score is calculated using formula (2). Representing components The relational integrity score is calculated using formula (3). The standard score representing the geometric integrity in the readyness standard fingerprint and , The standard score representing the completeness of attributes in a ready fingerprint is... , The standard score representing the relational completeness in the readyness standard fingerprint and , This represents the weighting coefficient of geometric completeness in similarity calculation, and is dimensionless. , This represents the weighting coefficient of attribute completeness in similarity calculation, and is dimensionless. , This represents the weighting coefficient of relation completeness in similarity calculation, and is dimensionless. And satisfy .

[0038] It should be noted that, due to , , The range of values ​​is , , The maximum weighted Euclidean distance within the square root is ,therefore The range of values ​​is When all three dimensions score 1 When all three dimensions score 0 .

[0039] Optionally, The value is 0.4. The value is 0.35. The value is 0.25. This value aligns with the weighting logic of the three types of relationships within the relational integrity in formula (3), where geometric representation, as the foundation of digital twin visualization, receives the highest weight; attribute information, as the basis for semantic analysis, receives the next highest weight; and the integrity of relational instances receives a relatively low weight. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to actual circumstances, and this embodiment of the invention does not impose specific limitations on this.

[0040] Furthermore, when At that time, components Components were identified as not ready, among which Represents the readiness threshold and is dimensionless. .

[0041] Optionally, The value is set to 0.85. This value strikes a balance between data quality controllability and component retention rate. When the threshold is set to 0.85, components with scores higher than 0.8 in all three dimensions can be filtered, while components with scores lower than 0.6 in any dimension will be judged as not ready. This effectively intercepts quality defects without over-filtering components. It is understood that those skilled in the art can adjust the threshold according to the actual situation. The value of is adjusted, but the embodiments of the present invention do not impose specific limitations on this.

[0042] Furthermore, for components determined to be incomplete... Retrieve components from the component type knowledge base A set of historical components of the same type is used to statistically analyze the distribution of attribute values ​​for each missing attribute. The attribute value with the highest probability of occurrence is used as the recommended completion value. The recommended completion value is then written into the BIM model after user confirmation. This mechanism utilizes historical statistical information of similar components for probabilistic recommendations, balancing the efficiency of automation with the accuracy of manual review.

[0043] In this embodiment of the invention, step S1 uses semantic fingerprint vectors to perform component-level granular quantitative evaluation and targeted repair of the data quality of each component in the BIM model, providing a quality-controllable data foundation for the subsequent scene-driven trimming in step S2. Compared with the existing method of directly converting the format of the BIM model, this method lacks the ability to quantitatively evaluate the quality of component-level data, and low-quality data will enter the digital twin model indiscriminately. However, this embodiment of the invention uses semantic fingerprint vectors to achieve multi-dimensional comprehensive evaluation, which can locate the dimension where the quality defects are located and provide targeted completion strategies, reducing the risk of low-quality data spreading to the digital twin model.

[0044] S2: Intent Graph Pruning and Enhancement: Based on the target application scenario, query the pre-built intent graph to extract the corresponding data requirement description. Based on the data requirement description, prune the minimized sub-model from the BIM model processed in step S1, which contains only the components and attributes required by the scenario. Perform targeted semantic enhancement on the minimized sub-model.

[0045] Specifically, an intent graph is a graph structure that organizes digital twin application scenarios into intent nodes, denoted as... ,in, Represents the set of intent nodes. This represents the set of edges connecting intent nodes. Each intent node... Description of requirements for associating a set of data ,in, Indicates the first One intention node, Indicates the intent node index and , Indicates the total number of intent nodes and , Representing a scene The required set of component types, Representing a scene The required set of attribute fields Representing a scene The required level of detail. Given a target application scenario, query the intent graph. Locate the corresponding intent node Extract its associated data requirement description .

[0046] It should be noted that the intent graph is predefined by domain experts based on the standardized classification of digital twin application scenarios. The edges between intent nodes represent data dependencies or inheritance relationships between scenarios. For example, the fire safety scenario can inherit the component type requirements of the building structure scenario.

[0047] Furthermore, based on the data requirement description The minimized sub-model is trimmed from the BIM model processed in step S1. The trimmed component set. The determination rule is shown in formula (5): in, This represents the set of components after trimming. Indicates the first Each component Representing components Type identifier, Data requirement description The set of component types specified in the document. This represents the component calculated by formula (4) in step S1. semantic fingerprint similarity, This represents the readiness threshold in step S1. This indicates that the set belongs to the operation. This represents the logical AND operation. The minimum submodel contains only... The set of attribute fields for each component The specified attribute.

[0048] It should be noted that the first condition in formula (5) The second condition is to remove scene-independent data at the component type level. At the quality level, the evaluation results of step S1 are used to ensure that all components entering the sub-model meet the data readiness requirements. The logical AND operation of the two conditions enables the sub-model to simultaneously meet the two objectives of minimization and quality control.

[0049] Furthermore, targeted semantic enhancement is performed on the minimized sub-model obtained from the trimming process. This involves supplementing the sub-model with attribute information required by the target application scenario but not included in the BIM model. Enhancement rules apply different strategies depending on the type of target application scenario, specifically including: supplementing the heat transfer coefficient and thermal inertia index of the building envelope for energy consumption analysis scenarios; marking evacuation distances and fire compartment information for fire safety scenarios; and supplementing rated power ranges and operating temperature ranges for equipment operation and maintenance scenarios. The data required for enhancement comes from building code databases, equipment product libraries, and industry standard parameter tables. (Scenario is defined.) The required set of enhancement attributes is The enhancement operation then minimizes each component in the submodel. Supplementary attributes ,in, Representing a scene The required set of enhanced attributes Representing components Existing attribute set, The set difference operation means removing elements contained in the second set from the first set.

[0050] Furthermore, when multiple different target application scenarios are specified for the same BIM model, each target application scenario shares the same BIM model processed in step S1. Steps S2 to S4 are executed independently for each target application scenario to generate the corresponding digital twin model. This design ensures that the quality assessment and completion in step S1 only needs to be performed once, and reusing the result across multiple scenarios can reduce computational redundancy.

[0051] In this embodiment of the invention, step S2, scene-aware pruning based on intent graphs, converges the BIM model, after quality completion in step S1, into a minimized sub-model oriented towards a specific application. Targeted semantic enhancement further supplements the scene-specific attributes required, providing semantically accurate and scalable input for multi-resolution coding in the subsequent step S3. Compared to the existing method of fully importing the BIM model, this method cannot organize data differently according to scene requirements and lacks a formal description mechanism for scene requirements. In contrast, this embodiment of the invention establishes an explicit mapping relationship between scene requirements and data structures through intent graphs, making the pruning process interpretable and reproducible.

[0052] S3: Semantic-geometric decoupling construction: Extract the building topology skeleton from the minimized sub-model output in step S2, construct the semantic topology layer with the building topology skeleton and component attribute information, independently encode the geometric data of each component into a geometric representation layer according to multi-level detail, and output a multi-resolution digital twin model in which the semantic topology layer and the geometric representation layer are decoupled.

[0053] Specifically, the building topology skeleton is extracted from the minimized sub-model output in step S2. ,in, Represents a set of nodes, including spatial nodes and component nodes. Represents a set of edges. Architectural topological relationships, i.e., edge types, include three categories: spatial connectivity represented by access through doors and windows. System topology represented by pipeline connections And the functional hierarchy represented by floors-areas-rooms. .

[0054] Furthermore, using the building topology framework Constructing a semantic topology layer from component attribute information The semantic topology layer is stored in a graph structure, with each node carrying a set of attributes for a component or space, and each edge carrying the type and semantic annotation of the topological relationship. The semantic topology layer is independent of the geometric data and contains only topological structure and attribute information.

[0055] Furthermore, the geometric data of each component is independently encoded into a geometric representation layer according to multiple levels of detail. The multi-level hierarchy of detail comprises three levels: the envelope level, represented by the smallest bounding box; the simplified block level, represented by simplified polyhedra of the main volumes; and the refined model level, expressed in complete 3D geometry. The loading level of each component is determined by weighting factors based on view distance, semantic importance, and task relevance. The overall hierarchy of detail score is then used. The calculation formulas are shown in formulas (6) and (7): in, Representing components The overall level of detail score and , This represents a three-factor linear weighted sum and , Representing components The viewing distance factor is normalized to the distance between the component and the current viewpoint. The values ​​are dimensionless, with closer intervals having higher values. Representing components The semantic importance factor and based on the components in the building topology skeleton The degree centrality of is determined by normalization and its range is . And it is dimensionless. Representing components Task-related factors and descriptions based on data requirements The level of detail specified for this component type in the target application scenario is normalized and its value range is [value range missing]. And it is dimensionless. The weighting coefficients of the sight distance factor are dimensionless and , The weight coefficients of the semantic importance factors are dimensionless and , The weight coefficients of the task-related factors are dimensionless and And satisfy , This represents the slope control parameter of the Sigmoid function, which is dimensionless and , The center offset parameter of the Sigmoid function is dimensionless and , This represents an exponential function with the natural constant as its base.

[0056] It should be noted that the denominator in formula (6) is ,because The denominator is always greater than 1, ensuring that the division is valid and .when Significantly higher than When the denominator approaches 1, it makes Approaching ,when Significantly lower than When the denominator is large, it makes Compressed. This characteristic allows detail level scores to retain key information in high-value regions and decay rapidly in low-value regions, thus achieving a smooth transition in level selection and avoiding drastic jumps in scores near boundaries. When all three factors are 0, , The lowest level of component loading; when all three factors are 1, , When the value approaches 1, the component is loaded at the highest level.

[0057] Optionally, The value is 0.3. The value is 0.3. The value is 0.4. The value is 8.0. The value is 0.5. (The above...) The basis for selecting the highest weight is that the task-related factor directly reflects the scenario's detailed requirements for components and should be the dominant factor in hierarchy selection. (The above...) A value of 8.0 gives the Sigmoid curve a moderately steep slope in the transition region, effectively distinguishing components at different levels within this transition area. The above... A value of 0.5 places the transition center at the midpoint of the scoring interval. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to actual circumstances, and this embodiment of the invention does not impose specific limitations in this regard.

[0058] Furthermore, according to The score maps the component to the corresponding level of detail: Load fine-grained model levels at time. Simplified block hierarchy is loaded at any time. Load the envelope level at time, where Represents the fine-level threshold and is dimensionless. , Represents the block level threshold and is dimensionless. .

[0059] Optionally, The value is 0.7. The value is 0.4. The above value will... The score interval is divided into three sub-intervals of approximately equal width, ensuring a relatively balanced distribution of components across the three levels and achieving a balance between rendering accuracy and computational overhead. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to actual circumstances, and this embodiment of the invention does not impose specific limitations in this regard.

[0060] Furthermore, the output multi-resolution digital twin model consists of a semantic topology layer. With geometric representation layer The two are linked by a unique identifier for each component, but their data organization is independent of each other.

[0061] In this embodiment of the invention, step S3 structurally decouples semantic information from geometric information. The semantic topology layer can perform semantic queries, attribute updates, and knowledge reasoning independently of geometric data, while the geometric representation layer can independently adjust the level of detail based on terminal performance and viewing distance without affecting the integrity of the semantic layer. This decoupling architecture provides a natural extension interface for injecting inference results into the semantic topology layer in step S4. Compared with the existing method of integrated storage of geometry and semantics, this method requires simultaneous processing of geometric data when performing semantic updates and reloading of the complete model when switching levels of detail, resulting in storage redundancy and response latency. In contrast, the decoupling architecture of this embodiment allows semantic updates and geometric rendering to be executed in parallel, which is beneficial for improving system response efficiency in multi-terminal collaborative access scenarios.

[0062] S4: Functional semantic inference injection: Extract the functional space semantic map from the building topology skeleton constructed in step S3, infer the implicit functional information not explicitly modeled in the BIM model based on the building function knowledge base, and inject the inference results into the semantic topology layer constructed in step S3.

[0063] Specifically, from the architectural topology framework Extracting functional space semantic graph ,in, This represents a collection with rooms or spaces as nodes, and each node is labeled with the space usage attribute. This represents a set whose edges are access relations. The functional space semantic graph contains... spatial nodes ,in, Indicates the first A spatial node, Represents the spatial node index and , Represents the total number of spatial nodes and spatial nodes The purpose attribute is The area is And the unit is square meters and .

[0064] It should be noted that the building function knowledge base is built based on national building design codes and industry standards. It encodes the domain knowledge correspondence between space use and functional requirements, and stores the per capita area standard, lighting power density standard, air conditioning cooling load index, noise level standard, and sound insulation and heat insulation requirement level lookup table corresponding to each space use type.

[0065] Furthermore, based on the correspondence between spatial uses and functional requirements in the building function knowledge base, and based on the extracted functional space semantic map... The implicit functional information of each spatial node is inferred. The inference of implicit functional information includes the following four types, and each type of inference is independent of the others.

[0066] In one possible implementation, depending on the space use and area Estimated personnel capacity Based on this, fire evacuation needs can be inferred. Personnel capacity. The calculation formula is shown in formula (8): in, Representing spatial nodes Inferred personnel capacity and unit is people and , Representing spatial nodes The area is expressed in square meters. This indicates the spatial usage in the building function knowledge base. The corresponding per capita area standard, and the unit is square meters per person. , Indicates the purpose of the space The corresponding peak occupancy rate coefficient is dimensionless and It also reflects the statistical probability of this type of space reaching full occupancy in actual use. This indicates a floor operation, which takes the largest integer not exceeding the input value. Because... The fact that the denominator is always greater than zero ensures that the division is valid.

[0067] Optionally, the office The value is 7 square meters per person, for the meeting room. The value is set at 2.5 square meters per person, for the office. The value is 0.7, for the conference room. The value is 0.9. It is understood that those skilled in the art can adjust the value of the above parameter according to actual building design specifications, and the embodiments of the present invention do not specifically limit this.

[0068] Furthermore, based on the inferred personnel capacity Further, based on fire safety regulations, we can infer the required evacuation width and number of evacuation exits for this space.

[0069] In one possible implementation, depending on the space use Inferring lighting design load and air conditioning design load Lighting design load The calculation formula is shown in formula (9): in, Representing spatial nodes The lighting design load is in watts. Representing spatial nodes The area is expressed in square meters. This indicates the spatial usage in the building function knowledge base. The corresponding standard value for lighting power density, and the unit is watts per square meter. , Representing spatial nodes The natural daylighting correction factor is dimensionless and The value is determined based on whether the space is adjacent to an exterior window and the window-to-wall ratio. When the space is adjacent to an exterior window and the window-to-wall ratio meets the lighting requirements, the value ranges from 0.6 to 0.8. When there is no exterior window, the value is 1.0.

[0070] Optionally, the office The value is 9 watts per square meter, for the corridor. The value is 5 watts per square meter, for the computer room. The value is set at 15 watts per square meter. Optional, For spaces with exterior windows and a window-to-wall ratio of not less than 0.3, the value is 0.7; for spaces without exterior windows, the value is 1.0. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to actual lighting design standards and building daylighting conditions, and this embodiment of the invention does not specifically limit this.

[0071] Furthermore, air conditioning design load The calculation formula is shown in formula (10): in, Representing spatial nodes The air conditioning design load is in watts. This indicates the spatial usage in the building function knowledge base. The corresponding air conditioning cooling load index benchmark value and the unit is watts per square meter. , Representing spatial nodes The orientation correction factor is dimensionless and ranges from -0.05 to 0.15. It is determined based on the influence of the exterior wall orientation on solar radiation heat gain, with values ​​ranging from 0.10 to 0.15 for west-facing walls and from -0.05 to 0 for north-facing walls. Representing spatial nodes The personnel density correction coefficient is dimensionless and ranges from 0 to 0.2. The personnel capacity is inferred from formula (8). With area The ratio of the actual population density Space Use Corresponding baseline personnel density The deviation between the two is determined by linearly mapping the actual personnel density to the range of 0 to 0.2 according to the excess ratio when the actual personnel density exceeds the benchmark personnel density, and taking 0 when the actual personnel density does not exceed the benchmark personnel density.

[0072] Optionally, the office The value is set at 120 watts per square meter for the conference room. The value is taken as 150 watts per square meter. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to actual air conditioning design standards, and the embodiments of the present invention do not specifically limit this.

[0073] It should be noted that formulas (9) and (10) respectively introduce corrections for spatial characteristics such as lighting conditions, orientation, and personnel density, so that the inference results can reflect the differences in actual working conditions of the space. Compared with the method of estimating the load by multiplying a single area index by the power density standard value, this method does not distinguish spatial characteristics such as lighting conditions and orientation, resulting in all spaces with the same area and purpose obtaining the same estimation results; while formulas (9) and (10) introduce multi-factor corrections so that the inference results can reflect the individual differences of the space and have higher engineering applicability.

[0074] In one possible implementation, based on the functional space semantic graph Accessibility in Infer ventilation paths and noise propagation chains. Ventilation paths are determined by... A breadth-first search is performed along the accessibility edges to identify connected paths from the intake space to the exhaust space. The noise propagation chain is determined by labeling the noise source levels of each spatial node and modeling its attenuation along the accessibility relationships. The propagation values ​​of noise in adjacent spaces are also considered. The calculation formula is shown in formula (11): in, Indicates noise from spatial nodes Propagation to adjacent spatial nodes The propagation value, expressed in decibels. Represents the index of adjacent spatial nodes and and , Representing spatial nodes The noise source level is specified in decibels and determined based on the typical noise levels corresponding to the space's use in the building function knowledge base. Representing spatial nodes With adjacent spatial nodes The acoustic transmission coefficient of the separating components is dimensionless and its value range is [value missing]. Furthermore, the smaller the value, the better the sound insulation performance, and the value is determined based on the material and structure of the partition components. This represents logarithmic operations with base 10. Representing spatial nodes With spatial nodes The additional attenuation value due to the propagation distance, in decibels. .because , Logarithmic operations are valid.

[0075] It should be noted that in formula (11) Equivalent to sound transmission loss in the field of acoustics, i.e. Sound insulation performance is the standard way to characterize sound insulation performance.

[0076] Optionally, between devices The value ranges from 75 to 85 decibels, in the office. The value ranges from 45 to 55 decibels. Optionally, for a 240 mm thick concrete wall... The value is 0.001, which is typical for lightweight partition walls. The value is 0.01. Optionally, The value of the parameter is determined based on the principle that the distance between the centroids of two spaces decreases by 6 dB with each doubling of the distance. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to actual circumstances, and this embodiment of the invention does not impose specific limitations on this.

[0077] In one possible implementation, the sound insulation or thermal insulation requirements of the partition structure are inferred based on the differences in the uses of adjacent spaces. Let the adjacent space nodes in the functional space semantic diagram be... and Their uses are respectively and Sound insulation requirements for partition components The calculation formula is shown in formula (12): in, Representing spatial nodes With spatial nodes The sound insulation requirements of the separating components are dimensionless integers. This represents the baseline sound insulation level value obtained by looking up a table in the building function knowledge base based on the combination of uses of adjacent spaces, and it is a dimensionless integer. Indicates based on spatial nodes Noise source level With spatial nodes Noise source level The difference between the noise levels determines the correction amount for the sound insulation requirement and is a dimensionless integer. When the absolute value of the noise level difference increases by 10 dB, the correction amount increases by 1 level.

[0078] Furthermore, the insulation requirements of the partition components. The calculation formula is shown in formula (13): in, Representing spatial nodes With spatial nodes The insulation requirement level of the separating components is a dimensionless integer. This represents the baseline thermal insulation rating value obtained from a table lookup based on the combination of uses of adjacent spaces in the building function knowledge base, and is a dimensionless integer. Indicates the purpose of the space The corresponding design temperature, in degrees Celsius. Indicates the purpose of the space The corresponding design temperature, in degrees Celsius. It represents the correction amount for insulation requirements determined based on the design temperature difference between two spaces, and is a dimensionless integer. The correction amount increases by 1 level for every 5 degrees Celsius increase in the absolute value of the temperature difference.

[0079] It should be noted that both formulas (12) and (13) employ a superposition structure of benchmark lookup values ​​and dynamic correction values, ensuring that the inference results both conform to the standard specifications and reflect the differences in specific working conditions. Benchmark lookup values and Predefined by the building function knowledge base based on the combination of space uses, with adjustments made accordingly. and The noise propagation value obtained from formula (11) and the actual difference between the space design temperature are dynamically adjusted. The function mapping relationship of the correction amount is a piecewise linear function, that is, the continuous difference interval is divided into several equal-width segments, and each segment corresponds to an integer level correction amount.

[0080] Optionally, when the office and meeting room are adjacent The value is 3, when the office and the equipment room are adjacent. The value is 5. Optionally, when the office is adjacent to the server room... The value is 4. Optionally, The value is 26 degrees Celsius for ordinary offices, 22 degrees Celsius for server rooms, and 18 degrees Celsius for warehouses. It is understood that those skilled in the art can adjust the values ​​of the above parameters according to actual building sound and heat insulation design specifications, and this embodiment of the invention does not specifically limit this.

[0081] Furthermore, confidence assessment and tiered injection are performed based on the above inference results. Each inference result is accompanied by a confidence score and inference path. Confidence Score The calculation formula is shown in formula (14): in, The confidence score of the inference result is dimensionless and , This represents the reliability factor of the data source and is dimensionless. The value is 1.0 when the source is a national standard, 0.9 when the source is an industry standard, and 0.7 when the source is empirical data. This represents the completeness factor of the input attributes and is dimensionless. is defined as the percentage of attributes already present in the input attributes required for inference. The length of the inference chain represents the number of inference steps from the input attribute to the inference result, and is a positive integer. When directly inferring from the table, the value is taken as 1. The decay control parameter representing the inference chain length is dimensionless and , This represents an exponential function with the natural constant as its base.

[0082] It should be noted that when hour Formula (14) degenerates into In other words, the confidence level of direct table lookup inference is determined solely by the reliability of the data source and the completeness of the input attributes. As the length of the inference chain increases... As the exponential decay factor increases, the confidence level gradually decreases, reflecting the risk of error accumulation during multi-step inference. Because... , , , It is always greater than zero.

[0083] Optionally, The value is set to 0.15. This value causes the confidence decay factor to become [value missing] for each additional step in the inference chain. This means a decrease in confidence of approximately 14%, striking a balance between encouraging multi-step reasoning and controlling the reliability of inference. It is understood that those skilled in the art can adjust the approach based on the specific circumstances. The value of is adjusted, but the embodiments of the present invention do not impose specific limitations on this.

[0084] Furthermore, the confidence score Not lower than the confirmation threshold The inference results are directly injected into the semantic topology layer. Those with confidence scores below the confirmation threshold The inference results are marked as pending confirmation and submitted to the user for review. After confirmation, they are injected into the semantic topology layer. This indicates that the threshold value is confirmed and is dimensionless. .

[0085] Optionally, The value is set to 0.75. This value allows for automatic injection of direct table lookup inferences based on national standards with complete input attributes, while inferences based on empirical data or involving two or more inference chains require user confirmation, thus striking a balance between automation efficiency and inference reliability. It is understood that those skilled in the art can adjust the values ​​according to the specific circumstances. The value of is adjusted, but the embodiments of the present invention do not impose specific limitations on this.

[0086] In this embodiment of the invention, step S4 extracts the functional space semantic graph based on the building topology skeleton constructed in step S3, systematically infers the implicit functional information not explicitly modeled in the BIM model, and injects the inference results, after confidence evaluation, into the semantic topology layer constructed in step S3. Since the semantic topology layer and the geometric representation layer remain decoupled, the addition of implicit functional information does not lead to redundant loading of geometric data, thus achieving a balance between semantic richness and system performance. Compared with the existing technology that relies solely on explicit data from the BIM model to construct a digital twin, this method lacks the ability to infer implicit functional information, and structured information such as space usage and topological relationships is not utilized. However, this embodiment of the invention, through the collaborative inference of the functional space semantic graph and the building function knowledge base, transforms structured information such as space usage and topological relationships into operable engineering decision parameters such as personnel capacity, equipment load, and sound and heat insulation requirements, providing an information support foundation for intelligent building operation and maintenance.

[0087] The above embodiments can be implemented, in whole or in part, by software, hardware, or any other combination. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium, an optical medium, or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0088] It should be understood that in various embodiments of the present invention, there is a sequential dependency of data transmission between the above steps S1 to S4. The output of step S1 serves as the input of step S2, the output of step S2 serves as the input of step S3, and the output of step S3 serves as the input of step S4. However, the specific implementation methods within each step can be adjusted according to the actual situation, and should not limit the implementation process of the embodiments of the present invention.

[0089] Those skilled in the art will recognize that the algorithmic steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. 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 implementations should not be considered beyond the scope of this invention.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A method for creating a digital twin model based on BIM, characterized in that, Includes the following steps: S1: Semantic fingerprint assessment and completion. For each component in the BIM model, an assessment score is calculated from three dimensions: geometric completeness, attribute completeness, and relational completeness, and these scores are combined into a semantic fingerprint vector. The semantic fingerprint vector is then compared with a pre-constructed readiness standard fingerprint. Components with a similarity score below the readiness threshold are identified as incomplete components, and their missing information is completed based on a component type knowledge base. The semantic fingerprint vector is a comprehensive feature vector obtained by quantifying the completeness of a component in the three dimensions of geometric representation, attribute definition, and relational instance. S2: Intent Graph Pruning and Enhancement. Based on the target application scenario, a pre-built intent graph is queried to extract the corresponding data requirement description. According to the data requirement description, a minimized sub-model containing only the required components and their attributes is pruned from the BIM model processed in step S1. Targeted semantic enhancement is then performed on the minimized sub-model. The intent graph is a graph structure that organizes the digital twin application scenario into intent nodes, each associated with the data requirement description. Targeted semantic enhancement involves supplementing the minimized sub-model with attribute information required by the target application scenario but not included in the BIM model. The BIM model in step S2 is the model after quality completion in step S1, ensuring that the pruning operation is performed based on a quality-controllable data foundation. S3: Semantic-geometric decoupling construction. An architectural topological skeleton is extracted from the minimized sub-model. A semantic topological layer is constructed using the architectural topological skeleton and component attribute information. The geometric data of each component is independently encoded into a geometric representation layer according to multi-level detail. A multi-resolution digital twin model, decoupled between the semantic topological layer and the geometric representation layer, is output. The architectural topological skeleton is a graph structure with spaces and components as nodes and architectural topological relationships as edges. Step S3 uses the minimized sub-model output in step S2 as input, ensuring that the extraction range of the topological structure has converged according to the scene requirements. S4: Functional semantic inference injection. A functional space semantic graph is extracted from the building topology skeleton. Based on the correspondence between space usage and functional requirements in the building function knowledge base, the implicit functional information not explicitly modeled in the BIM model is inferred from the space usage attributes of each node in the functional space semantic graph. The inference results are then injected into the semantic topology layer. The functional space semantic graph is a graph structure with rooms or spaces as nodes, accessibility relationships as edges, and each node labeled with a space usage attribute. The inference results from step S4 are injected into the semantic topology layer constructed in step S3, allowing the digital twin model to obtain supplementary implicit functional information while maintaining a geometrically and semantically decoupled structure.

2. The method for creating a BIM-based digital twin model according to claim 1, characterized in that, The geometric completeness is determined based on the shape descriptor matching score between the component's geometric representation and the corresponding level of detail standard template; the attribute completeness is determined based on the coverage of each required attribute in the standard attribute set corresponding to the component type; and the relational completeness is determined based on the instance completeness of the component in the IFC space containment relationship, component connection relationship, and system dependency relationship.

3. The method for creating a BIM-based digital twin model according to claim 1, characterized in that, The specific steps of supplementing the missing information of the unready component based on the component type knowledge base include: retrieving the attribute value distribution of historical components of the same type as the unready component from the component type knowledge base, using the attribute value with the highest probability of occurrence in the attribute value distribution as the recommended supplement value, and writing it into the BIM model after user confirmation.

4. The method for creating a BIM-based digital twin model according to claim 1, characterized in that, The data requirement description includes the set of component types, the set of attribute fields, and the level of detail required by the target application scenario; when performing pruning in step S2, components whose component types belong to the set of component types and whose similarity calculated in step S1 is not lower than the readiness threshold are retained.

5. The method for creating a BIM-based digital twin model according to claim 1, characterized in that, The targeted semantic enhancement applies different enhancement rules according to the type of the target application scenario. The enhancement rules include: supplementing the thermal parameters of the building envelope for energy consumption analysis scenarios, marking evacuation distances and fire compartment information for fire safety scenarios, and supplementing the range of equipment operating parameters for equipment operation and maintenance scenarios.

6. The method for creating a BIM-based digital twin model according to claim 1, characterized in that, The building topology includes spatial connectivity represented by doors and windows, system topology represented by pipeline connections, and functional hierarchy represented by floor-area-room containment levels.

7. The method for creating a BIM-based digital twin model according to claim 1, characterized in that, The multi-level detail hierarchy includes an envelope level, a simplified block level, and a fine model level; the loading level of each component is determined by weighting a view distance factor, a semantic importance factor, and a task association factor, wherein the task association factor is determined according to the level of detail specified for each component type based on the target application scenario described in the data requirement description.

8. The method for creating a BIM-based digital twin model according to claim 1, characterized in that, The inference of the implicit functional information includes: inferring the personnel capacity and fire evacuation requirements based on the space use and area; inferring the lighting and air conditioning design load based on the space use; inferring the ventilation path and noise propagation chain based on the accessibility relationship; and inferring the sound insulation or heat insulation requirements of the partition components based on the differences in the uses of adjacent spaces.

9. The method for creating a BIM-based digital twin model according to claim 8, characterized in that, The inference result includes a confidence score and an inference path; inference results with a confidence score not lower than the confirmation threshold are directly injected into the semantic topology layer, while inference results with a confidence score lower than the confirmation threshold are marked as pending confirmation and submitted to the user for review. After confirmation, they are injected into the semantic topology layer.

10. The method for creating a BIM-based digital twin model according to claim 1, characterized in that, When multiple different target application scenarios are specified for the same BIM model, each target application scenario shares the same BIM model processed by step S1. Steps S2 to S4 are executed independently for each target application scenario to generate the corresponding digital twin model.

Images