Fabricated building structure mechanical property optimization method and system based on BIM

Abstract

The invention provides a BIM-based fabricated building structural mechanical property optimization method and system, and relates to the technical field of model simulation, and the method comprises the steps: when a user carries out structural mechanical property optimization on a fabricated building based on a BIM model, capturing a user operation sequence in real time, and deconstructing an optimization decision path of the user; based on the optimization decision path, comparing with a preset optimization knowledge base, and identifying and positioning an optimization blind area possibly neglected by the user; wherein the optimized blind area at least comprises a coupling relation between thermal insulation material performance and structural node mechanical behaviors; and based on the optimization blind area, carrying out implicit heuristic intervention on the user, specifically, revealing the optimization potential in the optimization blind area through parameterized fine tuning, and feeding back a self-adaptive intervention process according to user cognition until the user is guided to autonomously complete optimization. The optimization precision of the mechanical property of the fabricated building structure is improved, the design efficiency and the user experience are remarkably improved, and meanwhile the safety and the energy-saving effect of the building are guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of model simulation, in particular to a prefabricated building structure mechanical property optimization method and system based on BIM. BACKGROUND

[0002] With the transformation of the construction industry towards digitization and intelligence, the technology based on building information model (BIM) is widely used in the design and optimization of prefabricated buildings.

[0003] Traditional structure mechanical property optimization methods rely on artificial experience or static analysis, which is difficult to respond to dynamic adjustments in the user design process in real time, and easily ignores the coupling relationship between the performance of thermal insulation materials and the mechanical behavior of structure nodes, etc. This leads to the possibility of local optimization or performance defects in the optimization results, affecting the safety and energy saving effect of the whole building. In addition, the existing optimization tools lack intelligent analysis and guidance of user decision paths, and the optimization efficiency is low, and users are difficult to find potential optimization space in complex parameter adjustment.

[0004] Therefore, there is an urgent need for a method that can capture user operations in real time, identify optimization blind spots, and guide users to optimize independently through implicit heuristic intervention, in order to improve the optimization accuracy and efficiency of prefabricated building structure mechanical properties. This method needs to combine the dynamic characteristics of the BIM model, fully tap the interactive potential of user decisions and pre-set knowledge base, and improve user experience and optimization effect through adaptive intervention, providing technical support for the intelligent design of prefabricated buildings. SUMMARY

[0005] The present application aims to provide a prefabricated building structure mechanical property optimization method and system based on BIM to solve the problems pointed out in the background art.

[0006] In the first aspect, the present application provides a prefabricated building structure mechanical property optimization method based on BIM, comprising:

[0007] When a user optimizes the structure mechanical properties of a prefabricated building based on a BIM model, the user's operation sequence is captured in real time, and the optimization decision path is deconstructed;

[0008] Based on the optimization decision path, the optimization blind spots that the user may ignore are identified and located by comparing with the pre-set optimization knowledge base; wherein the optimization blind spots at least include the coupling relationship between the performance of thermal insulation materials and the mechanical behavior of structure nodes;

[0009] Based on the optimization blind spots, implicit heuristic intervention is carried out on the user, specifically: the optimization potential in the optimization blind spots is revealed through parameterized fine-tuning, and the intervention process is adaptively intervened according to user cognitive feedback until the user independently completes the optimization.

[0010] Secondly, the present invention provides a BIM-based prefabricated building structure mechanical performance optimization system, comprising:

[0011] The capture and deconstruction module is used to capture the user's operation sequence in real time and deconstruct its optimization decision path when the user optimizes the structural mechanical performance of prefabricated buildings based on the BIM model.

[0012] The comparison and identification module is used to compare the optimized decision path with a pre-set optimization knowledge base to identify and locate optimization blind spots that users may overlook; among them, optimization blind spots include at least the coupling relationship between the performance of thermal insulation materials and the mechanical behavior of structural nodes;

[0013] The implicit heuristic intervention module is used to provide implicit heuristic intervention to users based on optimization blind spots. Specifically, it reveals the optimization potential in the optimization blind spots through parameterized fine-tuning and adaptively intervenes based on user cognitive feedback until it guides users to complete the optimization autonomously.

[0014] The present invention has achieved the following beneficial effects:

[0015] By capturing user operation sequences in real time and deconstructing the optimization decision path, it is possible to accurately identify blind spots that users overlook during the optimization process, especially the coupling relationship between the performance of thermal insulation materials and the mechanical behavior of structural nodes. Through implicit heuristic intervention and parametric fine-tuning, it dynamically reveals the optimization potential and adaptively guides users to complete the optimization. This not only improves the optimization accuracy of the mechanical performance of prefabricated building structures, but also significantly improves design efficiency and user experience, while ensuring the safety and energy-saving effect of buildings.

[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0017] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0019] Figure 1 This is a schematic diagram of a BIM-based method for optimizing the mechanical performance of prefabricated building structures in an embodiment of the present invention.

[0020] Figure 2This is a schematic diagram of a BIM-based prefabricated building structure mechanical performance optimization system according to an embodiment of the present invention. Detailed Implementation

[0021] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0022] The research and development approach of this application addresses the shortcomings of traditional BIM optimization methods, which lack dynamic response and intelligent guidance. It proposes to capture user operation sequences in real time, analyze decision paths by combining them with a pre-set optimization knowledge base, identify blind spots such as the coupling between insulation materials and structural nodes, and achieve adaptive guidance through parametric fine-tuning and implicit heuristic intervention.

[0023] Figure 1 This application provides a flowchart of a BIM-based method for optimizing the mechanical performance of prefabricated building structures, as shown in the embodiments below. Figure 1 As shown, the method includes:

[0024] S1. When a user optimizes the structural mechanical performance of a prefabricated building based on the BIM model, the user's operation sequence is captured in real time, and their optimization decision path is deconstructed. Step S1 specifically includes the following sub-steps:

[0025] S11. Real-time monitoring and capture of user interaction events with the BIM model to form a user operation sequence sorted by timestamp; wherein, the user operation sequence includes at least the following: modification operations on component geometric properties, material parameters and node connection methods.

[0026] Real-time monitoring and capture of user interaction events with the BIM model refers to continuously monitoring user actions within the BIM modeling environment through a software interface, generating a user operation sequence containing timestamps, operation types, and operation content. A user operation sequence is defined as a collection of operation records recorded chronologically. Each record includes an operation timestamp, the operation object (e.g., component, material, or node), the operation type (e.g., modify, add, delete), and specific operation parameters (e.g., geometric dimensions, material strength, node stiffness). Component geometric attributes refer to the component's dimensional parameters (e.g., length, width, thickness), obtained through the geometric editing module of the BIM software, in millimeters. Material parameters refer to the component's physical and mechanical properties (e.g., elastic modulus, compressive strength), extracted from the BIM model's material property library, in MPa and kN / m, respectively. 2Node connection methods refer to the connection forms between components (such as hinged or rigid connections), obtained through the node definition module of the BIM model, and are represented by connection type and stiffness value (unit: kN·m / rad). Real-time monitoring captures user operation events through the BIM software's API (such as Revit API). Event data includes the operation trigger time (accurate to milliseconds), the operation object ID, and the modified parameter value. After capture, the event data is sorted in ascending order by timestamp, forming a user operation sequence, which is stored in a memory buffer for subsequent analysis. To ensure data integrity, the system sets the event capture frequency (at least 100 times per second) to avoid missing fast operations, and the buffer size is dynamically adjusted according to the operation frequency (e.g., number of operations per minute × size of a single record, approximately 100kB).

[0027] Here is a specific implementation example:

[0028] In prefabricated building design scenarios, designers use Autodesk Revit for BIM modeling to optimize the mechanical properties of a precast concrete frame structure. The system captures user actions in real time through the Revit API's DocumentChanged event listener. Whenever a user modifies component geometry properties (e.g., adjusting the beam's cross-sectional height from 500mm to 600mm), material parameters (e.g., changing concrete strength from C30 to C40, increasing compressive strength from 20.1MPa to 26.7MPa), or node connection methods (e.g., changing beam-column connection from hinged to rigid, adjusting stiffness from 0kN·m / rad to 10000kN·m / rad), the listener records the operation timestamp (e.g., 14:32:45.123), object ID (e.g., beam B001), operation type (e.g., modifying geometry), and specific parameter changes. The listener runs at 100Hz to ensure the capture of rapid, continuous operations, with data temporarily stored in a memory buffer (pre-allocated 200kB, sufficient to store approximately 6000 operations per minute). For example, if a user continuously modifies the cross-sectional height of three beams, the strength of two materials, and the connection method of one node within 10 seconds, the system generates six operation records, sorted by timestamp as the sequence: [14:32:45.123, Beam B001, Modify geometry, Height 500mm→600mm], [14:32:45.456, Beam B002, Modify geometry, Height 500mm→550mm], etc. After the sequence is stored, a buffer verification mechanism (comparing the number of records with the expected number of operations) ensures no data loss for subsequent analysis steps.

[0029] S12. Analyze the temporal relationship and logical connection between each operation in the user operation sequence to deconstruct the decision steps taken by the user to achieve a specific mechanical performance goal.

[0030] Analyzing the temporal and logical relationships of user operation sequences involves using algorithms to analyze the temporal order and causal dependencies of each operation in the sequence, extracting the decision steps taken by the user to optimize mechanical performance (such as increasing structural stiffness or reducing displacement). Temporal relationships are defined as the relative time differences between operations, calculated by comparing operation timestamps, in milliseconds. Logical relationships are defined as the functional dependencies between operations, such as adjusting material parameters to maintain strength after modifying component geometric properties, or changing node connection methods to optimize force transmission, based on mechanical constraints (such as force balance and stiffness matching) of the BIM model. Decision steps are defined as combinations of operations to achieve specific mechanical performance goals; for example, "increasing beam stiffness" may include "increasing beam cross-section" and "increasing material strength." The analysis employs a temporal relationship mining algorithm, and the construction process is as follows: First, the operation sequence is initialized as a directed temporal graph, with nodes representing operation records and edges representing time differences (weighted by timestamp differences). Then, based on mechanical constraint rules (such as the positive correlation between component geometry and material strength), the logical association strength between operations is calculated, with the strength value ranging from [0, 1]. The angle between the operation parameter vectors (composed of geometry, material, and node parameters) is calculated using cosine similarity. Finally, a clustering algorithm (based on DBSCAN, with a density threshold of 0.5 and a distance threshold of 100ms) clusters strongly associated operations into decision steps, outputting a step sequence (e.g., "Step 1: Adjust beam cross-section → Step 2: Modify material"). Timestamps are directly extracted from the operation sequence, mechanical constraint rules are derived from the mechanical simulation module of the BIM model (e.g., RevitStructure), and the logical association strength can also be calculated using a predefined mechanical parameter mapping table.

[0031] Continuing with the Revit scenario above, the user operation sequence obtained from S11 contains 6 records, such as: [14:32:45.123, Beam B001, modify geometry, height 500mm→600mm], [14:32:45.456, Beam B001, modify material, strength C30→C40], etc. The temporal association mining algorithm first constructs a directed temporal graph, with the 6 operation records as nodes and the edge weights representing the time difference (e.g., the time difference from 45.123 to 45.456 is 333ms). The algorithm calculates the logical association strength based on the mechanical constraint rules of the BIM model (e.g., an increase in beam cross-section height requires matching with higher strength materials). For example, the cosine similarity between the vectors of the geometry modification and material modification of beam B001 ([height 600mm, strength 26.7MPa]) is 0.85, indicating a strong association. The DBSCAN algorithm (density threshold 0.5, distance threshold 100ms) clusters six operations into three decision steps: Step 1 (adjustment of beam B001 section and material, aiming to increase stiffness), Step 2 (adjustment of beam B002 section, aiming to optimize self-weight), and Step 3 (adjustment of rigid connection at node N001, aiming to enhance force transmission). The mechanical objectives of each step are verified through the structural analysis module of the BIM model. For example, the stiffness of beam B001 is increased from 5000 kN·m. 2 Increased to 7500 kN·m 2 This aligns with the user's optimization intent. The algorithm runs in a Python environment, using NumPy to calculate vector similarity and SciPy to implement DBSCAN, ensuring efficient processing of at least 1000 operation sequences (approximately 1 second). The output decision step sequence is stored in memory for use by S13.

[0032] S13. Based on the deconstructed decision steps, generate a structured optimized decision path.

[0033] Generating a structured optimization decision path refers to organizing the decision steps deconstructed from S12 into an ordered path description that includes mechanical objectives and operational content, used to characterize the overall logic of the user's optimization behavior. An optimization decision path is defined as a set of decision steps arranged in chronological order. Each step includes a mechanical objective (e.g., increasing stiffness, reducing displacement), operational content (e.g., modifying the beam cross-section), and influencing parameters (e.g., stiffness increment, in kN·m). 2The path structuring uses a directed acyclic graph (DAG), where nodes represent decision steps and edges represent logical dependencies between steps (e.g., after adjusting the beam cross-section in step 1, step 2 requires adjusting the material to match the stiffness). The generation process includes: First, extracting the decision step sequence of S12 and mapping it to DAG nodes, labeling each node with a mechanical objective (obtained from the mechanical simulation results of the BIM model, such as stiffness change). Then, constructing DAG edges based on the logical association strength between steps (cosine similarity calculated in S12), with weights equal to the association strength (range [0, 1]). Finally, using a topology sorting algorithm (such as the Kahn algorithm) to sort the DAG, ensuring that the path reflects the temporal sequence and logical dependencies of the operations. The mechanical objective is extracted from the structural analysis module of the BIM model (e.g., beam stiffness change from 5000 kN·m). 2 Up to 7500 kN·m 2 The logical association strength follows the results from S12, and the DAG edge weights are filtered by a threshold (weights greater than 0.7 are retained). The paths are stored as a graph structure in memory for comparison in subsequent steps.

[0034] For example, in a Revit scene, the three decision steps output by S12 (adjusting the section and material of beam B001, adjusting the section of beam B002, and adjusting the rigid connection of node N001) serve as inputs. The system constructs a DAG, with the three steps as nodes, and the nodes labeled with mechanical objectives (e.g., the objective for step 1 is "increase the stiffness of beam B001 to 7500 kN·m"). 2 The path is obtained through stiffness analysis of Revit Structure. Based on the logical association strength of S12 (e.g., the association strength between step 1 and step 2 is 0.75, and between step 2 and step 3 is 0.80), DAG edges are generated, and the filtering threshold is set to 0.7 to retain strongly dependent edges. The Kahn algorithm performs topological sorting on the DAG and outputs the optimization decision path: [Step 1: Beam B001 section 600mm + material C40 → Step 2: Beam B002 section 550mm → Step 3: Node N001 rigid connection]. The path reflects the user's logic of first optimizing the stiffness of beam B001, then adjusting the self-weight of beam B002, and finally enhancing the force transmission at the nodes. Path generation is implemented using the NetworkX library in a Python environment. DAG nodes and edges are stored as graph objects, and processing 100 steps takes approximately 0.5 seconds. The path is recorded through Revit's log module for comparison with step S2, ensuring that the user's optimization intent is fully expressed.

[0035] S2. Based on the optimized decision path, compare it with the pre-built optimization knowledge base to identify and locate optimization blind spots that users may overlook; among them, optimization blind spots include at least the coupling relationship between the performance of thermal insulation materials and the mechanical behavior of structural nodes. Step S2 specifically includes the following sub-steps:

[0036] S21. Construct an abstract network representation of a pre-built optimized knowledge base; wherein, the nodes in the abstract network representation represent the design entities and performance indicators, and the edges represent the mechanical coupling and constraint relationships between them.

[0037] The abstract network representation of constructing a pre-built optimization knowledge base refers to organizing the mechanical optimization knowledge of prefabricated buildings into a network structure. Nodes represent design entities (such as beams, columns, and nodes) and performance indicators (such as stiffness and displacement), while edges represent mechanical couplings (such as the positive correlation between beam cross-section and stiffness) or constraint relationships (such as the compatibility between node stiffness and component strength). The abstract network representation is defined as a weighted directed graph, where nodes contain entity parameters (geometric dimensions, material strength, units of mm and MPa) and performance indicator values ​​(stiffness kN·m). 2 The displacement is 600 mm, and the edge weight is the coupling strength (range [0, 1], calculated in advance through mechanical simulation). The construction process includes: First, extracting the design entity and performance indicators from the mechanical analysis module of the BIM model (such as SAP2000) to generate a node set. For example, the node of beam B001 contains parameters [section height 600 mm, strength 26.7 MPa] and indicators [stiffness 7500 kN·m]. 2 Then, finite element analysis (FEA) is used to calculate the mechanical coupling between entities. For example, a 10% increase in beam cross-section height leads to a 15% increase in stiffness, with a coupling strength of 0.9 (based on FEA sensitivity analysis). Constraints are defined according to engineering specifications (such as the "Code for Design of Concrete Structures"), for example, the node stiffness must be less than twice the member stiffness. Finally, the network is stored as an adjacency matrix, and edge weights are normalized (maximum values ​​are normalized to 1). Entity parameters are extracted from the BIM model, performance indicators are calculated through FEA simulation, coupling strength is based on FEA sensitivity analysis, and constraint relationships are queried from the specification database.

[0038] For example, in the design of prefabricated concrete frames, the system uses SAP2000 to import beam B001, column C001, and node N001 from a Revit model, generating a node set: Beam B001 [section 600mm × 300mm, strength 26.7MPa, stiffness 7500kN·m] 2 Column C001 [section 400mm×400mm, strength 26.7MPa, displacement 5mm], node N001 [rigid connection, stiffness 10000kN·m / rad]. Using SAP2000's FEA module, it was calculated that when the height of beam B001 section increases by 10% (600mm→660mm), the stiffness increases by 15% (7500kN·m). 2 →8625kN·m 2The network is constructed using the NetworkX library, with nodes and edges stored as a weighted directed graph. A constraint edge is generated (beam B001 section → stiffness, weight 0.9). The constraint relationships are based on the "Code for Design of Concrete Structures" GB50010, verifying that the stiffness of node N001 (10000 kN·m / rad) is less than twice the stiffness of beam B001 (15000 kN·m / rad), and a constraint edge is generated (node ​​N001 → beam B001, weight 1.0). The network is built using the NetworkX library, with nodes and edges stored as a weighted directed graph. Generating a network of 100 nodes takes approximately 2 seconds. Coupling strength is calculated in batches using FEA (approximately 0.1 seconds per simulation). Constraint relationships are queried from a pre-defined code database to ensure the network fully reflects the mechanical relationships for use in S22 mapping.

[0039] S22. Map the optimized decision path to an abstract network representation and perform a semantic matching operation to generate a path coverage graph and a relation-ignoring weight graph.

[0040] Mapping the optimized decision path to an abstract network representation involves comparing the decision path generated in S13 with the knowledge base network in S21, identifying the nodes and edges covered by the path, and generating a path coverage graph and a relation neglect weight graph. The path coverage graph is defined as a subgraph of the abstract network, containing nodes (e.g., components, performance indicators) and edges (e.g., coupling relationships) involved in the decision path. The relation neglect weight graph is defined as a complement graph of the abstract network, where nodes are uncovered entities or indicators, and edge weights represent the degree of neglect (range [0, 1], based on the coupling strength of uncovered edges). The mapping process includes: First, mapping each step of the decision path to network nodes and edges; for example, adjusting the beam cross-section corresponds to beam nodes and cross-section-stiffness edges. Then, using a semantic matching algorithm (based on graph embedding vector similarity), the matching degree between path nodes and network nodes is calculated. The matching degree is calculated using the cosine similarity of node parameter vectors (geometry, material, indicators), with a threshold of 0.8. After matching, the covered nodes and edges are extracted to generate the path coverage graph, and the remaining uncovered nodes and edges are used to generate the relation neglect weight graph, where the neglect weight is equal to the coupling strength of the uncovered edges. The decision path is obtained from S13, network nodes and edges are extracted from S21, and the matching degree is calculated by vector operation in NumPy. The weights of the uncovered edges are directly inherited without regard to the weights.

[0041] For example, in the above scenario, the decision path of S13 (adjustment of beam B001 section and material → adjustment of beam B002 section → rigid connection of node N001) is mapped to the network of S21. Step 1 of the path (beam B001 section 600mm + material C40) corresponds to the network node [beam B001, section 600mm, strength 26.7MPa, stiffness 7500kN·m]. 2The semantic matching algorithm uses Node2Vec (implemented on NetworkX, training 100 nodes takes about 1 second) to generate node embedding vectors, calculates the cosine similarity between the beam B001 step vector and the network node vector (e.g., 0.92, exceeding the threshold of 0.8), and confirms a match. The path coverage graph contains 3 nodes (beam B001, beam B002, node N001) and 2 edges (section → stiffness, node → stiffness). The relationship-ignoring weight graph contains uncovered nodes (e.g., column C001) and edges (e.g., column C001 displacement → beam B001 stiffness, weight 0.85). The weight is ignored by directly taking the coupling strength of the uncovered edge (0.85). The algorithm runs in Python and takes about 0.8 seconds to process the mapping of 100 nodes. The generated coverage graph and the ignored weight graph are stored as NetworkX subgraphs for S23 analysis to ensure accurate identification of the coverage of user optimization behavior.

[0042] S23. Optimize the coverage completeness of the decision path on the network structure based on path coverage graph analysis, and optimize the attention of the decision path to specific mechanical coupling relationships based on relation neglect weight graph analysis.

[0043] The analysis of the coverage completeness and attention of the optimization decision path refers to evaluating the comprehensiveness of user optimization and the degree of attention to key mechanical coupling relationships by quantifying the coverage ratio and relationship of the path coverage graph to the knowledge base network, ignoring the weight distribution of the weight graph. Coverage completeness is defined as the proportion of the number of nodes and edges in the path coverage graph to the total number of nodes and edges in the knowledge base network, ranging from [0, 1], calculated by direct counting (number of covered nodes / total number of nodes, number of covered edges / total number of edges). Attention is defined as the proportion of the sum of edge weights of a specific mechanical coupling relationship (e.g., the coupling between thermal insulation material and the mechanical behavior of nodes) in the path coverage graph to the sum of the corresponding edge weights in the knowledge base network, ranging from [0, 1]. The analysis process includes: first, calculating the node coverage rate (node ​​ratio) and edge coverage rate (edge ​​ratio) of the path coverage graph, and then taking a weighted average (node ​​weight 0.6, edge weight 0.4) to obtain the coverage completeness. Secondly, edges with specific coupling relationships (such as insulation material → node stiffness) are extracted from the relational neglect weight graph, and the ratio of their total weights to the total weights of the corresponding edges in the knowledge base network is calculated to obtain the attention level. The coverage graph and neglect weight graph are obtained from S22, the number of nodes and edges is counted through NetworkX, the coupling relationship weights are inherited from S21, and the attention threshold (e.g., 0.7) is set based on engineering experience.

[0044] For example, in the above scenario, the knowledge base network contains 100 nodes (50 entities, 50 indicators) and 200 edges, and the path coverage graph contains 3 nodes and 2 edges. The coverage completeness is calculated as follows: node coverage = 3 / 100 = 0.03, edge coverage = 2 / 200 = 0.01, and the weighted average is (0.6 × 0.03 + 0.4 × 0.01 = 0.022). Attention analysis focuses on the coupling between the insulation material and the mechanical behavior of the nodes. The total weight of relevant edges in the knowledge base network (e.g., insulation material → node stiffness, weight 0.8) is 2.4 (3 edges). Ignoring the total weight of uncovered edges of this type in the weighted graph, the attention score is 1.6. Therefore, attention score = 1 - (1.6 / 2.4) ≈ 0.333. The calculation uses NetworkX's graph attribute interface, and processing 100 nodes takes approximately 0.3 seconds. The results show that the coverage completeness of 0.022 is far below the threshold of 0.5, and the attention level of 0.333 is also below the threshold of 0.7, indicating that the user's optimization ignored a large number of mechanical relationships and insulation-node couplings. The analysis results are stored as a set of indicators in memory for S24 to locate optimization blind spots, ensuring that the evaluation accurately reflects the shortcomings of the user's optimization.

[0045] S24. Network areas with coverage completeness lower than the preset coverage completeness threshold and attention lower than the corresponding attention threshold are identified as optimization blind spots.

[0046] Optimization blind spots refer to knowledge base network regions with coverage completeness and attention levels below a certain threshold, marked as optimization areas that users may overlook. An optimization blind spot is defined as a network subgraph not covered by the path coverage graph but containing key mechanical coupling relationships, such as the coupling region between insulation material performance and node mechanical behavior. The coverage completeness threshold (e.g., 0.5) is set based on engineering experience, indicating that the optimization path must cover at least 50% of network nodes and edges. The attention threshold (e.g., 0.7) indicates that the coverage ratio of key coupling relationships must reach 70%. The localization process includes: First, comparing the coverage completeness of S23 with the threshold, filtering network regions below 0.5. Then, extracting coupling relationship subgraphs with attention levels below 0.7 from the neglect weight graph (e.g., insulation material → node stiffness). Finally, merging the filtering results to generate an optimization blind spot subgraph, where nodes represent uncovered entities and indicators, and edges represent unaddressed coupling relationships. Coverage completeness and attention levels are obtained from S23, the thresholds are set based on the "Technical Specification for Prefabricated Concrete Structures," and the subgraph is generated using NetworkX's subgraph extraction algorithm.

[0047] For example, in the scenario above, S23 calculates a coverage completeness of 0.022 (threshold 0.5) and an attention score of 0.333 (threshold 0.7). The system filters out uncovered areas in the knowledge base network and extracts a coupled subgraph of the neglect weight graph showing the relationship between insulation material and node mechanical behavior (containing nodes [insulation material, thermal conductivity 0.04 W / (m·K)] and [node N002, stiffness 8000 kN·m / rad], and edge [insulation material → node stiffness, weight 0.8]). Subgraph extraction uses NetworkX's subgraph method, processing 100 nodes takes approximately 0.2 seconds. Optimization blind spots are marked as subgraphs containing insulation material and node N002, reflecting the user's neglect of the impact of insulation performance on node mechanics (e.g., increased insulation layer thickness may reduce node stiffness). The blind spot subgraph is stored as a NetworkX object for S3 intervention, ensuring accurate location of insufficient optimization areas by the user.

[0048] S3. Based on the optimization blind spot, provide implicit heuristic intervention to the user. Specifically, this involves revealing the optimization potential within the blind spot through parameterized fine-tuning, and adaptively adjusting the intervention process based on user cognitive feedback until the user is guided to complete the optimization autonomously. Step S3 includes the following sub-steps:

[0049] S31. Conduct a global sensitivity analysis on the key design parameters corresponding to the optimization blind zone, and screen out the set of parameters that have a significant impact on the overall mechanical performance.

[0050] Global sensitivity analysis refers to quantifying the contribution of design parameters in the blind zone to mechanical performance and screening the set of parameters that significantly affect overall performance (such as stiffness and displacement). Key design parameters are defined as entity parameters in the blind zone subgraph (such as the thermal conductivity of insulation materials and nodal stiffness), with units of W / (m·K) and kN·m / rad, respectively. Overall mechanical performance refers to structural stiffness (kN·m). 2 Indicators such as maximum displacement (mm) were obtained through FEA simulation of the BIM model. Sensitivity analysis adopted the Sobol method, and the construction process was as follows: First, the parameter space was defined (e.g., thermal conductivity [0.03, 0.05] W / (m·K), nodal stiffness [5000, 15000] kN·m / rad). Then, at least 1000 sets of parameter samples were generated (using Latin hypercube sampling), and the mechanical properties of each set of samples were simulated using SAP2000 to calculate the Sobol first-order exponent (the contribution of the parameter to the performance variance, ranging from [0, 1]). The screening criterion was that the first-order exponent was greater than 0.2, indicating a significant impact. The parameter set contained the screened parameters and their ranges. Blind zone parameters were extracted from the S24 subgraph, performance indicators were obtained through FEA simulation, and the Sobol exponent was calculated using the SALib library in Python.

[0051] For example, in the scenario described above, the optimization blind zone includes the thermal conductivity of the insulation material (0.04 W / (m·K)) and the stiffness of node N002 (8000 kN·m / rad). The Sobol analysis defines the parameter space as [0.03, 0.05] W / (m·K) and [5000, 15000] kN·m / rad, generating 1000 sets of samples using the SALib library (approximately 0.5 seconds). Each set of samples is simulated using SAP2000, calculating the structural stiffness (e.g., as the thermal conductivity increases from 0.04 to 0.05, the stiffness increases from 7500 kN·m). 2 Reduced to 7200 kN·m 2 The Sobol first-order exponent analysis showed that thermal conductivity contributed 0.25 to stiffness and nodal stiffness contributed 0.35, both exceeding the threshold of 0.2. The selected parameter set consisted of [thermal conductivity, nodal stiffness], with ranges of [0.03, 0.05] and [5000, 15000], respectively. The analysis was run in a Python environment, with approximately 10 seconds required for 1000 simulations. The parameter set was stored as an in-memory list for adjustment by the S32 algorithm, ensuring that the selected parameters had a significant impact on mechanical properties.

[0052] S32. Apply limited, reversible, automated adjustments to the parameter set in accordance with engineering specifications, and generate corresponding exploratory optimization schemes.

[0053] Applying finite-range, reversible automated adjustments to a parameter set refers to automatically modifying the parameters selected by S31 under engineering specification constraints to generate exploratory optimization schemes to reveal potential blind spots. The finite range is defined as the parameter adjustment range not exceeding 10% of the specification's allowable range (e.g., the material parameter variation range specified in the "Code for Design of Concrete Structures"). Reversibility means the adjustment can be restored through the BIM model's undo function. An exploratory optimization scheme is defined as a scheme containing adjusted parameters and predicted performance, with performance calculated through FEA simulation. The adjustment process includes: first, based on the parameter range of S31 (e.g., thermal conductivity [0.03, 0.05]), increasing the parameters in increments of 5% (e.g., 0.04 → 0.042). Then, simulating the adjusted mechanical properties (e.g., stiffness, displacement) using SAP2000. The adjustment complies with specifications (e.g., the thermal conductivity still meets energy-saving specifications after adjustment). After the scheme is generated, it is stored as a parameter-performance pair. The parameter range is obtained from S31, specification constraints are queried from the database, and performance is calculated through FEA simulation.

[0054] For example, in the scenario above, the adjustment range of the parameter set [thermal conductivity 0.04 W / (m·K), nodal stiffness 8000 kN·m / rad] is set to 5%. The system adjusts the thermal conductivity to 0.042 W / (m·K) and the nodal stiffness to 8400 kN·m / rad, which complies with the "Standard for Energy-Saving Design of Buildings" GB50189 and the "Code for Design of Concrete Structures". SAP2000 simulation shows that the structural stiffness increases to 7600 kN·m after adjustment. 2 The displacement was reduced to 4.8 mm. An exploratory optimization scheme was proposed with a thermal conductivity of 0.042, a node stiffness of 8400, and a stiffness of 7600 kN·m. 2 [Displacement 4.8mm], automatically applied via Revit's API (approximately 0.2 seconds), and undoable. The scheme is stored as an in-memory object for S33 monitoring, ensuring adjustments reveal blind spot potential and comply with specifications.

[0055] S33. Monitor users' interactive responses to the exploratory optimization plan in real time and update users' cognitive status accordingly.

[0056] Real-time monitoring of user interaction responses refers to capturing user actions on S32 exploratory optimization schemes via BIM software APIs, updating user cognitive status to assess their acceptance of the schemes. Interaction response is defined as a user's actions of viewing, modifying, or ignoring the scheme, acquired through the Revit API's ElementSelection event, recording the operation type and timestamp. User cognitive status is defined as the user's level of attention to the optimization scheme, quantified as a one-dimensional score (range [0, 1], 0 indicating complete ignoring, 1 indicating complete adoption), calculated through interaction frequency and operation depth (frequency = number of operations / monitoring period, depth = number of modified parameters). The monitoring process includes: first, capturing user actions (such as clicking on scheme parameters) in real time and recording them as interaction events. Then, calculating the cognitive status score: frequency (number of operations per minute) accounts for 0.6 weight, and depth (number of modified parameters / total number of parameters) accounts for 0.4 weight. The updated cognitive status is stored as an in-memory variable. Interaction events are captured via API, and frequency and depth are calculated through counting and proportion.

[0057] For example, in a Revit scenario, S32's tentative solution [thermal conductivity 0.042, node stiffness 8400] is displayed through the Revit interface. The system uses the Revit API to monitor user actions, capturing 3 views (frequency = 3 / 5 = 0.6 times / minute) and 1 modification of the thermal conductivity to 0.041 (depth = 1 / 2 = 0.5) within 5 minutes. The cognitive state score = 0.6 × 0.6 + 0.4 × 0.5 = 0.56, indicating moderate attention. The monitoring runs in Python, processing 100 interactions in approximately 0.1 seconds. The cognitive state is updated to a memory variable of 0.56 for S34 to determine, ensuring accurate reflection of the user's response to the solution.

[0058] S34. When it is confirmed that the user is in a state of continuous neglect based on the updated user cognitive state, progressively increase the visibility of the exploratory optimization scheme until the user's adoption behavior or active interaction signal is captured.

[0059] Enhancing the visual saliency of the exploratory optimization scheme refers to guiding users to focus on the scheme by adjusting the BIM interface display (such as highlighting, pop-ups) when the user's cognitive state is below a threshold (e.g., 0.7), until adoption behavior (modification of scheme parameters) or active interaction signal (viewing the scheme more than 3 times) is detected. Visual saliency is defined as the visual attractiveness of interface elements, quantified as color contrast (range [0, 1], based on RGB difference) and pop-up frequency (times / minute). Progressive enhancement refers to increasing saliency over time, such as increasing contrast by 10% or adding 1 pop-up per minute. Ignoring the state is confirmed by comparing the cognitive state score with the threshold of 0.7; enhancement is triggered when the score is below the threshold. The enhancement process includes: first, adjusting the display color of the scheme parameters (RGB from [200, 200, 200] to [255, 0, 0]) and the pop-up frequency (from 0 to 2 times / minute). Then, monitoring user interaction to detect adoption (modification of parameters) or active interaction (views ≥ 3 times). Cognitive state is obtained from S33, contrast is calculated using RGB values, and pop-up frequency is set using a timer.

[0060] For example, in the scenario described above, S33's cognitive state is 0.56, below the threshold of 0.7, triggering saliency enhancement. The system adjusts the display color of the scheme parameters (thermal conductivity 0.042, node stiffness 8400) from RGB[200, 200, 200] to [255, 0, 0] (contrast improvement of 20%) via the Revit API, setting the pop-up frequency to 1 time / minute, displaying "It is recommended to adjust the thermal conductivity to optimize stiffness." After 2 minutes, 3 user views are detected (active interaction signal), and the enhancement stops. The processing runs in the Revit environment, with interface adjustments taking approximately 0.1 seconds. Ultimately, active interaction is captured, and the scheme is stored as a log to ensure that users are guided to focus on blind spot optimization.

[0061] This invention captures user operation sequences in real time and deconstructs optimization decision paths, enabling it to accurately identify blind spots overlooked by users during the optimization process, especially the coupling relationship between the performance of thermal insulation materials and the mechanical behavior of structural nodes. Through implicit heuristic intervention and parametric fine-tuning, it dynamically reveals optimization potential and adaptively guides users to complete optimization, which not only improves the optimization accuracy of the mechanical performance of prefabricated building structures, but also significantly improves design efficiency and user experience, while ensuring the safety and energy-saving effect of buildings.

[0062] Step S1 lays the foundation for subsequent intelligent analysis and guidance by capturing user operation sequences in real time and deconstructing optimization decision paths. Sub-step S11 captures user interaction events with the BIM model through high-frequency monitoring and API, forming a timestamp-sorted operation sequence. This accurately records operation time, objects, and parameter changes, ensuring data integrity and real-time performance, and preventing the omission of rapid operations. Sub-step S12 utilizes temporal correlation mining algorithms and mechanical constraint rules to analyze the temporal relationships and logical connections between operations, extracting user decision steps and clearly revealing optimization intentions. Sub-step S13 generates structured optimization decision paths through directed acyclic graphs and topological sorting, integrating scattered operations into an ordered logical expression for easy comparison with the knowledge base. This systematic operation capture and path deconstruction method not only improves the accuracy of user behavior analysis but also provides a reliable data foundation for identifying optimization blind spots, effectively supporting subsequent intelligent intervention.

[0063] Step S2, by comparing with a pre-built optimization knowledge base, accurately identifies blind spots in the user's optimization process, especially the coupling relationship between insulation materials and the mechanical behavior of structural nodes, providing crucial evidence for intelligent guidance. Sub-step S21 constructs an abstract network representation, organizing design entities, performance indicators, and mechanical coupling relationships into a weighted directed graph, clearly expressing complex mechanical constraints and providing a structured knowledge foundation for subsequent matching. Sub-step S22 uses a semantic matching algorithm to map the optimization decision path to the knowledge base network, generating a path coverage graph and a relationship neglect weight graph, accurately identifying the mechanical relationships covered and ignored by the user. Sub-step S23 quantifies coverage completeness and attention, assessing the comprehensiveness of the user's optimization and the degree of attention paid to key coupling relationships, revealing deficiencies. Sub-step S24 uses threshold filtering to locate optimization blind spots, identifying uncovered key areas. This series of steps, through systematic network analysis and quantitative evaluation, ensures accurate blind spot location, provides clear targets for subsequent intervention, and improves the pertinence and effectiveness of the optimization scheme.

[0064] Step S3, through implicit heuristic intervention and parametric fine-tuning, dynamically reveals the potential of optimization blind spots and adaptively guides users to complete optimization, significantly improving design efficiency and optimization effectiveness. Sub-step S31 employs global sensitivity analysis to screen parameters that significantly impact mechanical performance, providing high-priority adjustment targets for optimization and reducing the cost of blind adjustments. Sub-step S32 applies limited-amplitude, reversible automated adjustments under engineering specification constraints, generating exploratory optimization schemes. Performance changes are predicted through FEA simulation, ensuring the safety and feasibility of adjustments. Sub-step S33 monitors user interaction responses to the schemes in real time, updates cognitive state scores, accurately assesses user acceptance, and provides a basis for subsequent guidance. Sub-step S34 progressively enhances visualization saliency, effectively attracting attention when users ignore the schemes until adoption or proactive interaction is triggered. This adaptive intervention mechanism, through dynamic adjustment and user feedback loops, not only increases user attention and adoption rates for blind spot optimization but also enhances the intelligence and user experience of the BIM optimization process, ensuring a comprehensive improvement in building mechanical performance and energy efficiency.

[0065] In some embodiments, the BIM-based method for optimizing the structural mechanical performance of prefabricated buildings further includes:

[0066] S4. When a user rejects an implicit heuristic intervention, the historical spatiotemporal scope used to analyze the reasons for the rejection is inferred based on the context in which the user's rejection occurred. Step S4 specifically includes the following sub-steps:

[0067] S41. Extract the key operation timestamp sequence related to the optimization blind spot from the user operation sequence, and combine it with the version iteration log of the BIM model to construct the operation impact propagation chain ending with the rejection behavior.

[0068] Extracting key operation timestamp sequences related to optimization blind spots from user operation sequences and combining them with BIM model version iteration logs, constructing an operation impact propagation chain ending with rejection behavior, involves analyzing user operation behavior in the BIM environment, filtering key operations related to optimization blind spots (such as the coupling of insulation material and structural node mechanical behavior), and combining this with model version change records to construct a directed propagation chain reflecting the causal relationship of operations, used to trace the context of rejection behavior. The key operation timestamp sequence is defined as a set of user operation records related to optimization blind spots. Each record includes an operation timestamp (accurate to milliseconds), the operation object (such as component, material, node), the operation type (such as modification, ignore), and parameter values ​​(such as thermal conductivity, stiffness). Optimization blind spots are obtained from S24 and include entities and mechanical relationships that have not been optimized by the user (such as the thermal conductivity of insulation materials and node stiffness). Version iteration logs refer to historical snapshots of the BIM model during the design process, recording the components, parameters, and time of each modification. These logs are extracted from the version control module of BIM software (such as Revit's DesignHistory), and are measured in milliseconds (timestamps) and specific parameter values ​​(such as thermal conductivity W / (m·K)). The operation impact propagation chain is defined as a directed graph, where nodes are key operations, edges represent causal relationships between operations, and edge weights are the correlation between time intervals (in milliseconds) and changes in mechanical parameters (range [0, 1], calculated using cosine similarity of parameter vectors). The construction process includes: First, filtering relevant operations from the user operation sequence in S11 based on entities and indicators (such as insulation materials and node stiffness) in the optimization blind zone to generate a timestamp sequence. Second, combining the version iteration logs, verifying the actual impact of operations on the model (such as whether thermal conductivity adjustments are effective). Then, using timestamps and parameter changes, a directed graph is constructed, where nodes are operation records, edges are determined by time differences and parameter correlations, and rejected behaviors are used as endpoint nodes. The timestamp sequence is captured through BIM software APIs (such as Revit API), the logs are extracted from the version control module, and the correlation is calculated using vector similarity using NumPy.

[0069] Here is a specific implementation example:

[0070] In the optimization scenario of a prefabricated concrete frame structure, the user manipulates the BIM model using Autodesk Revit. S24 identifies the optimization blind spot as the coupling relationship between the insulation material (thermal conductivity 0.04 W / (m·K)) and node N002 (stiffness 8000 kN·m / rad). After intervention in S34, the user refuses to adjust the thermal conductivity. The system uses the DocumentChanged event listener of the Revit API to filter operations related to the insulation material and nodes from the operation sequence of S11 (containing 6000 records, approximately 100kB), generating a timestamp sequence, for example: [14:32:45.123, Insulation material, View, Thermal conductivity 0.04], [14:32:46.456, Node N002, Ignore, Stiffness 8000]. Combined with Revit's DesignHistory module, version iteration logs are extracted, confirming that these operations did not cause changes to model parameters (e.g., the thermal conductivity remained at 0.04). The system build process influences the propagation chain. A directed graph is generated using the NetworkX library, with nodes representing 5 key operations (e.g., checking insulation material, ignoring node stiffness). Edge weights are determined by time difference (e.g., 333ms) and parameter correlation (cosine similarity between thermal conductivity and stiffness vectors, 0.82). Rejection behavior (ignoring thermal conductivity adjustment) serves as the endpoint node. The graph contains 5 nodes and 4 edges, with a generation time of approximately 0.3 seconds (1000 records). Correlation calculations are based on NumPy, processing 100 operations in approximately 0.1 seconds. The propagation chain is stored as an in-memory NetworkX object to ensure accurate tracing of the context of rejection behavior for S42 analysis.

[0071] S42. Based on the semantic similarity and time decay weight of each operation node in the operation influence propagation chain, and by introducing the coupling strength factor of the structural mechanical parameter change between nodes, calculate the contribution of each operation node to the rejection behavior.

[0072] Based on the semantic similarity and time decay weight of each operation node in the operation influence propagation chain, and introducing the coupling strength factor of structural mechanical parameter changes between nodes, the contribution of each operation node to the rejection behavior is calculated. This means evaluating the degree of influence of each operation on the user's rejection behavior by quantifying the semantic association, temporal proximity, and mechanical parameter coupling between operations. Semantic similarity is defined as the similarity of the parameter vectors of operation nodes, ranging from [0, 1], calculated using cosine similarity. The parameter vectors include the geometry, material, or node properties of the operation object (e.g., thermal conductivity 0.04 W / (m·K), stiffness 8000 kN·m / rad). The time decay weight is defined as a weight decreasing factor that increases with the time interval, based on the exponential decay function exp(-λΔt), where λ is the decay rate (set to 0.01, unit 1 / second), and Δt is the difference between the operation timestamp and the rejection behavior timestamp (unit second). The coupling strength factor is defined as the degree of mutual influence of changes in mechanical parameters between operations, ranging from [0, 1]. It is calculated using finite element analysis (FEA) to determine the contribution of parameter changes to mechanical properties (such as stiffness), for example, the impact of changes in thermal conductivity on node stiffness. The contribution degree is defined as the comprehensive impact of the operating node on the rejection behavior, ranging from [0, 1]. The calculation formula is: Contribution degree = 0.4 × semantic similarity + 0.3 × time decay weight + 0.3 × coupling strength factor. The calculation process includes: First, extracting the parameter vectors of each node in the S41 propagation chain and calculating the semantic similarity with the rejection behavior node (using NumPy). Second, calculating the time decay weight based on the timestamp difference (using a Python math library). Then, analyzing the mechanical coupling strength of parameter changes between nodes using FEA (such as SAP2000), for example, an increase in thermal conductivity of 0.01 W / (m·K) leading to a 5% decrease in stiffness. Finally, the three weights are fused to calculate the contribution degree. The propagation chain is obtained from S41, and the FEA data is extracted from the mechanical simulation module of the BIM model.

[0073] For example, in the Revit scenario above, the propagation chain of S41 contains 5 nodes, such as: [14:32:45.123, insulation material, view, thermal conductivity 0.04], [14:32:46.456, node N002, ignore, stiffness 8000], and the rejection behavior node is [14:32:47.789, insulation material, ignore, thermal conductivity 0.04]. The system extracts the node parameter vector (e.g., [thermal conductivity 0.04, stiffness 8000]), and uses NumPy to calculate the cosine similarity with the rejection behavior node. For example, the semantic similarity of node 1 is 0.90. The time decay weight is based on exp(-0.01×Δt), where Δt is 2.666 seconds (47.789-45.123), and the weight is 0.974. The coupling strength factor was simulated using SAP2000 to analyze the coupling between a thermal conductivity of 0.04 and a node stiffness of 8000. For example, an increase in thermal conductivity of 0.01 leads to a 5% decrease in stiffness, resulting in a coupling strength of 0.85 (based on FEA sensitivity analysis). The contribution was calculated as: 0.4 × 0.90 + 0.3 × 0.974 + 0.3 × 0.85 ≈ 0.907. The contributions of the five nodes were [0.907, 0.850, 0.820, 0.780, 0.750], with a calculation time of approximately 0.5 seconds (100 nodes). Each FEA simulation took approximately 0.1 seconds, and the coupling strength was processed in batches. The contribution values ​​were stored in an in-memory array for S43 filtering to ensure accurate quantification of the impact of the operation on the rejection behavior.

[0074] S43. Define the time interval and model space range covered by the set of operation nodes whose contribution exceeds the preset contribution threshold as the historical spatiotemporal range.

[0075] The historical spatiotemporal range is defined as the time interval and model space range covered by the set of operation nodes whose contribution exceeds a preset contribution threshold. This means identifying the time periods and BIM model space regions strongly correlated with rejection behavior by filtering high-contribution operation nodes, thus forming the context for analyzing the reasons for rejection. The contribution threshold is defined as 0.8, based on engineering experience, indicating that the operation's impact on the rejection behavior must reach 80% significance. The time interval is defined as the timestamp range of high-contribution nodes, from the earliest to the latest timestamp (in seconds). The model space range is defined as the BIM model entities (such as beams, nodes, and insulation materials) involved in the high-contribution nodes and their geometric boundaries (in millimeters, extracted from the BIM model's geometry module). The historical spatiotemporal range is defined as the Cartesian product of the time interval and the spatial range, representing the context of the rejection behavior. The filtering process includes: First, obtaining the contribution list from S42 and filtering nodes with a contribution greater than 0.8. Then, extracting the timestamps of these nodes, calculating the minimum and maximum timestamps to form the time interval. Next, extracting the model entity IDs of the nodes and querying the geometric boundaries (such as the boundary box of insulation materials) through the Revit API. Finally, the time interval and spatial range are merged to form the historical spatiotemporal range. The contribution value is obtained from S42, the timestamp and entity ID are extracted from the propagation chain of S41, and the geometric boundary is obtained through the geometric query interface of the BIM software.

[0076] For example, in the above scenario, the contribution calculated by S42 is [0.907, 0.850, 0.820, 0.780, 0.750], with a threshold of 0.8, filtering out 3 nodes: [14:32:45.123, insulation material, view, thermal conductivity 0.04], [14:32:46.456, node N002, ignore, stiffness 8000], [14:32:47.789, insulation material, ignore, thermal conductivity 0.04]. The time interval is [45.123, 47.789] seconds (approximately 2.666 seconds). The spatial range is queried using the Revit API's Element.GetBoundingBox method, including the insulation material (boundary box [0, 0, 0] - [1000, 1000, 50] mm) and node N002 (position [500, 500, 0] mm). The historical spatiotemporal range is defined as a combination of time [45.123, 47.789] seconds and space [insulation material, node N002]. Filtering and querying are performed using Python and the Revit API, processing 10 nodes in approximately 0.2 seconds. The spatiotemporal range is stored as an in-memory object for S5 analysis, ensuring accurate definition of the context of rejection behavior.

[0077] S5. Based on historical interaction data recorded within a historical timeframe, analyze the reasons why users reject implicit heuristic interventions. Step S5 specifically includes the following sub-steps:

[0078] S51. Within the historical time and space range, extract the user's response delay duration, modification operation frequency, and parameter rollback behavior to the exploratory optimization scheme, and use them as interaction history data.

[0079] Within a historical spatiotemporal scope, extracting user response delay duration, modification operation frequency, and parameter rollback behavior for exploratory optimization solutions, and using this as interaction history data, involves analyzing user interaction behavior with the S32 exploratory optimization solution within the spatiotemporal scope defined in S43. This quantifies user response speed, operation frequency, and reversal tendency, forming a dataset for analyzing rejection reasons. Response delay duration is defined as the time difference between the solution display and the first interaction, in seconds, calculated using the operation timestamp captured by the BIM software API. Modification operation frequency is defined as the number of times solution parameters are modified per unit time (per minute), in times / minute, counted using the API's event counter. Parameter rollback behavior is defined as the user's operation of restoring solution parameters to their pre-adjustment state, identified as a Boolean value (1 for rollback, 0 for no rollback), determined by comparing parameter values ​​before and after the operation. Interaction history data is defined as a set of records containing response delay, operation frequency, and rollback behavior. The extraction process includes: First, based on the spatiotemporal scope (time interval and spatial range) of S43, filtering operations related to the exploratory solution in the S11 operation sequence. Secondly, calculate the response latency for each operation (the difference between the timestamp displayed in the scheme and the operation timestamp). Then, count the number of modification operations within one minute and compare parameter values ​​to determine rollback behavior. Finally, compile the data into an interaction history. Timestamps and operations are obtained through the Revit API's ElementSelection event, and rollback behavior is determined by parameter comparison.

[0080] For example, in the above scenario, the spatiotemporal range of S43 is time [45.123, 47.789] seconds, space [insulation material, node N002], and the tentative solution is [thermal conductivity 0.042, node stiffness 8400]. The system filters the operation sequence through Revit API and obtains 3 related operations: [14:32:45.456, insulation material, view, thermal conductivity 0.042], [14:32:46.789, insulation material, modify, thermal conductivity 0.041], [14:32:47.789, insulation material, modify, thermal conductivity 0.04]. The response latency is 0.333 seconds (45.456-45.123). The modification operation frequency is 2 times / 2.666 seconds = 45 times / minute. The rollback behavior detects that the thermal conductivity has recovered from 0.042 to 0.04, and marks it as 1. The interaction history data is {[Response latency: 0.333 seconds], [Operation frequency: 45 times / minute], [Rollback: 1]}. Extraction was performed using the Revit API and Python, processing 10 operations in approximately 0.1 seconds. The data is stored as an in-memory list for S52 analysis to ensure accurate capture of user interactions with the solution.

[0081] S52. Construct a user decision-making behavior graph based on historical interaction data, and use a graph neural network to generate feature vectors of user decision-making patterns.

[0082] Constructing a user decision-making behavior graph based on historical interaction data and generating feature vectors of user decision-making patterns using a graph neural network (GNN) involves organizing the S51 interaction data into a graph structure to reflect user decision-making patterns and extracting high-dimensional feature vectors using GNN for subsequent rejection reason matching. The user decision-making behavior graph is defined as a weighted directed graph, where nodes represent interaction operations (including response latency, operation frequency, and rollback behavior), edges represent the temporal and semantic relationships between operations, and edge weights are time difference (in seconds) and parameter similarity (range [0, 1], calculated using cosine similarity). The GNN model is defined as a multi-layer neural network based on the GraphSAGE algorithm. The input is the node features of the graph (response latency, operation frequency, rollback value) and the adjacency matrix, and the output is a feature vector for each node (dimension set to 64). The GNN construction process includes: first, initializing the node feature vector to [response latency, operation frequency, rollback value] and normalizing it to [0, 1]. Then, constructing edges based on timestamp differences and parameter similarity, with weights of 0.5 × time weight + 0.5 × similarity. The GNN was trained using the PyTorchGeometric library, with a 3-layer GraphSAGE layer. Each layer aggregated neighbor information, and the activation function was ReLU. Training lasted 100 epochs (approximately 10 seconds per node). The generation process included: inputting the graph, and the GNN outputting a 64-dimensional feature vector representing the user's decision-making pattern. Interaction data was obtained from an S51 microcontroller, and similarity was calculated using NumPy. The GNN was run on a GPU (e.g., GPU).

[0083] It runs on an NVIDIA RTX 3060.

[0084] For example, in the above scenario, the S51's interaction history data {[response latency: 0.333 seconds], [operation frequency: 45 times / minute], [rollback: 1]} generates a graph. Each node represents three operations, and the normalized feature vectors are [0.333 / 10, 45 / 300, 1] = [0.0333, 0.15, 1]. Edges are weighted based on time difference (e.g., 1.333 seconds) and parameter similarity (cosine similarity of thermal conductivity vectors, 0.95), with a weight of 0.5 × exp(-0.01 × 1.333) + 0.5 × 0.95 ≈ 0.968. The graph is constructed using NetworkX, with 3 nodes and 3 edges, taking approximately 0.1 seconds to generate. The GNN is based on PyTorchGeometric's GraphSAGE, a 3-layer structure, with each layer aggregating 2-hop neighbors, trained for 100 epochs (approximately 5 seconds, GPU accelerated). After inputting the graph, the system outputs three 64-dimensional feature vectors, representing the user's decision-making patterns after quick viewing, modification, and rollback. These vectors are stored as in-memory tensors for S53 matching, ensuring accurate extraction of decision features.

[0085] S53. Perform multi-dimensional matching between the feature vector and the typical rejection pattern template in the pre-set optimization knowledge base, and identify the target cause based on the matching results; wherein, the target cause includes at least one of the following: insufficient understanding of the coupling relationship between thermal insulation materials and structural nodes, simplified decision-making strategies adopted due to time pressure, and lack of trust in automated adjustment based on historical experience.

[0086] The process involves multi-dimensional matching of feature vectors with typical rejection pattern templates in a pre-built optimized knowledge base to identify target reasons. This means determining the reasons for user refusal to intervene, such as insufficient understanding, time pressure, or lack of trust, by comparing the S52 GNN feature vectors with the rejection pattern templates in the knowledge base. Typical rejection pattern templates are defined as a pre-built set of feature vectors. Each type of rejection reason (insufficient understanding, time pressure, lack of trust) corresponds to a set of 64-dimensional vectors, generated statistically from historical cases and stored in the optimized knowledge base. Matching uses multi-dimensional cosine similarity to calculate the similarity between the feature vector and each template, ranging from [0, 1]. The template with the highest similarity is selected as the target reason. The matching process includes: First, extracting template vectors from the S21 knowledge base, such as the insufficient understanding template [quick viewing, low-frequency modification, high rollback rate]. Then, calculating the cosine similarity between the S52 feature vector and the template (using NumPy). Finally, selecting templates with a similarity greater than 0.8 and mapping them to the reasons: insufficient understanding (quick viewing but rollback), time pressure (high-frequency operation without modification), and lack of trust (ignoring intervention). The template is retrieved from the knowledge base, and the similarity is calculated using NumPy. The threshold of 0.8 is based on engineering experience.

[0087] For example, in the scenario described above, the feature vector of S52 (64-dimensional, representing quick viewing and rollback after modification) is compared with the templates in the S21 knowledge base. The knowledge base contains three types of templates: insufficient awareness [0.05, 0.2, 0.9, ...], time pressure [0.8, 0.9, 0.1, ...], and lack of trust [0.1, 0.1, 0.1, ...]. Using NumPy to calculate cosine similarity yields [0.92, 0.65, 0.55]. The template similarity for insufficient awareness (0.92 > 0.8) confirms this as the target reason. The matching process takes approximately 0.1 seconds (100 templates). The results indicate that the user refused intervention due to insufficient awareness of the coupling relationship between the insulation material and the nodes. The reason is stored as a memory label for S6 planning, ensuring accurate identification of the root cause of rejection.

[0088] S6. Based on the objective reasons, plan the expected cognitive path for users to re-understand the optimization potential, and set a maximum time limit for this understanding. Step S6 specifically includes the following sub-steps:

[0089] S61. Based on the type of the target cause, retrieve the cognitive evolution trajectory of similar cases from the optimized knowledge base and extract the sequence of key cognitive turning points.

[0090] Based on the type of the target reason, the cognitive evolution trajectory of similar cases is retrieved from the optimized knowledge base to extract key cognitive turning point sequences. This refers to querying the cognitive change path of users from rejection to acceptance of intervention in historical cases based on the rejection reasons identified by S53 (such as insufficient cognition), extracting key behavioral nodes, and forming a reference sequence to guide users to re-cognize. The cognitive evolution trajectory is defined as the sequence of operations from user rejection to acceptance of intervention, recorded as timestamps, operation types (such as viewing, modifying), and parameter changes (such as thermal conductivity adjustment). Key cognitive turning points are defined as operation nodes in the trajectory that lead to cognitive change, such as the modification operation of the first adoption of intervention. The optimized knowledge base contains a trajectory dataset of historical cases, stored through an abstract network representation of S21, where nodes represent operations and edges represent time and semantic associations. The retrieval process includes: First, filtering the knowledge base based on the target reason (such as insufficient cognition) to extract similar cases (similarity > 0.8, based on cosine similarity of trajectory vectors). Then, using the DBSCAN algorithm (density threshold 0.5, distance threshold 100ms) to cluster high-frequency operation nodes in the trajectory, identifying them as turning points. The turning point sequence includes operation time, type, and parameters. The case was retrieved from the S21 knowledge base, and the similarity was calculated using NumPy. DBSCAN was implemented using SciPy.

[0091] For example, in the scenario above, S53 identifies the cause as insufficient understanding. The system retrieves 100 cases from the S21 knowledge base. The trajectory vectors (operation type, parameters) have a similarity > 0.8 with the insufficient understanding template, resulting in 10 selected cases. Each case contains approximately 50 operation records, such as [viewing thermal conductivity, modifying stiffness, adopting intervention]. DBSCAN (density threshold 0.5, distance threshold 100ms) clusters three inflection points: [viewing insulation material description, 14:33:00], [modifying thermal conductivity to 0.042, 14:33:10], and [confirming stiffness improvement, 14:33:20]. The sequence reflects the user's cognitive shift through learning material knowledge, attempting adjustments, and verifying performance. Retrieval and clustering are performed using Python and SciPy, processing 100 cases in approximately 1 second. The inflection point sequences are stored as an in-memory list for S62 planning, ensuring the provision of effective cognitive references.

[0092] S62. Based on the learning ability indicators and historical operation proficiency in user profile data, a cognitive dynamics model is used to simulate the rational cognitive process from a state of rejection to autonomous acceptance of intervention, generating a staged cognitive path that includes a sequence of cognitive triggering events, an expected cognitive deepening operation chain, and cognitive verification nodes, which serves as the expected cognitive path.

[0093] Based on user profile data, including learning ability indicators and historical operational proficiency, a cognitive dynamics model is used to simulate the rational cognitive process from rejection to acceptance of intervention, generating a phased cognitive path. This involves quantifying user learning ability and operational experience, simulating the dynamics of cognitive change, and generating an operational sequence to guide users to accept intervention. The learning ability indicator is defined as the speed at which a user understands new knowledge, ranging from [0, 1], and calculated by normalizing historical learning task completion time (in seconds). Historical operational proficiency is defined as the user's operational efficiency in the BIM environment, ranging from [0, 1], and calculated by operation frequency (times / minute) and error rate (erroneous operations / total operations). The cognitive dynamics model is defined as a state transition model, with states representing cognitive stages (rejection, attention, attempt, acceptance), and transition probabilities based on learning ability and proficiency, constructed as a Markov chain. Model construction includes: initializing the state transition matrix with a transition probability of 0.4 × learning ability + 0.6 × proficiency; simulating 1000 transitions to generate a path from rejection to acceptance, including trigger events (viewing prompts), in-depth operations (modifying parameters), and verification nodes (confirming performance improvements). The path generation process includes: First, learning ability and proficiency are extracted from user profiles (through S11 operation sequence statistics). Then, a Markov chain is run to output high-probability paths. Each path consists of three stages, with 5-10 operations per stage. User profiles are extracted from BIM logs, and the model is implemented using Python.

[0094] For example, in the above scenario, the user profile shows a learning ability of 0.7 (based on an average completion time of 600 seconds for 10 learning tasks) and proficiency of 0.8 (operation frequency of 2 times / minute, error rate of 0.1%). The cognitive dynamics model uses Python to construct a Markov chain, with the transition matrix initialized to 4×4 (rejection, attention, attempt, acceptance), and the transition probability as 0.4×0.7+0.6×0.8=0.76 (rejection → attention). Simulating 1000 times (approximately 1 second), the generated paths are: [Rejection → View Prompt, 14:33:00], [Attention → Modify Thermal Conductivity to 0.042, 14:33:10], [Attempt → Verify Stiffness 7600kN·m] 2 [14:33:20], [Accept → Confirm Plan, 14:33:30]. The path contains 10 operations in 3 stages (trigger, refine, verify). The path is stored as a memory sequence for use by S63 to ensure that the simulation is reasonable and can guide user cognition.

[0095] S63. Based on the total estimated time of each stage in the expected cognitive path, and by introducing a risk adjustment factor that is dynamically adjusted according to the critical impact of the optimization blind zone on the structural mechanical performance, the latest cognitive time limit is set.

[0096] Based on the estimated total time consumed at each stage of the expected cognitive path, a risk adjustment factor is introduced to set the latest cognitive deadline. This refers to determining the latest time the user must complete the cognitive process by estimating the execution time of the S62 path and considering the impact of the optimization blind zone on mechanical performance. The estimated total time consumed is defined as the sum of the time consumed by each operation in the path, in seconds, estimated using the average time consumed by historical operations (statistics from S11). The risk adjustment factor is defined as the severity of the impact of the optimization blind zone on structural performance (such as stiffness and displacement), ranging from [0, 1], calculated through FEA simulation to account for performance losses due to changes in blind zone parameters (e.g., a 10% reduction in stiffness). The latest cognitive deadline is defined as the total time consumed multiplied by the adjustment factor, in seconds. The setting process includes: First, calculating the average time consumed by operations in the S62 path (e.g., viewing for 1 second, modifying for 2 seconds). Then, simulating the impact of the blind zone using SAP2000, for example, a thermal conductivity of 0.04 leading to an 8% stiffness loss, with a factor of 0.8. The deadline is time consumed × factor. The time consumed is extracted from S11, and the factor is calculated through FEA.

[0097] For example, in the scenario above, the S62 path contains 10 operations (5 view, 3 modify, 2 verify), with an average time of [1, 2, 3] seconds, totaling 5 × 1 + 3 × 2 + 2 × 3 = 17 seconds. SAP2000 simulation shows that an unoptimized thermal conductivity of 0.04 leads to an 8% reduction in stiffness, with a risk adjustment factor of 0.8. The time limit is 17 × 0.8 = 13.6 seconds. The calculation uses Python, and the FEA simulation takes approximately 0.1 seconds. The time limit is stored as a memory variable for S7 monitoring to ensure a reasonable cognitive timeframe is set.

[0098] S7. Monitor the deviation between the user's subsequent optimization decision path and the expected cognitive path, and trigger cognitive enhancement intervention when the deviation exceeds the limit or the latest cognitive time limit is reached. Step S7 specifically includes the following sub-steps:

[0099] S71. Monitor the incremental optimization decision path after user rejection in real time and extract the operation feature sequence.

[0100] Real-time monitoring of the incremental optimization decision path after user rejection and extraction of operational feature sequences refers to continuously capturing subsequent user actions in the BIM environment to generate feature sequences reflecting optimization behavior for comparison with the expected cognitive path. The incremental optimization decision path is defined as the newly generated operational sequence after user rejection of intervention, including timestamps, operation types, and parameter values. The operational feature sequence is defined as a quantitative representation of the operation, including operation type encoding (view = 1, modify = 2, etc.), parameter vectors (e.g., thermal conductivity 0.042), and timestamps (in seconds). The monitoring process includes: First, capturing operations in real time via the Revit API's DocumentChanged event, filtering records related to optimization blind spots. Second, extracting operational features, including type encoding (through a predefined mapping table), parameter vectors (extracted from the BIM model), and timestamps (obtained from the API). The feature sequences are sorted by timestamps and stored as an in-memory list. Operations are captured via the API, and feature extraction is performed using Python.

[0101] For example, in the scenario above, after the user refuses to intervene, the system captures three new operations via the Revit API: [14:33:01.123, Insulation material, View, Thermal conductivity 0.04], [14:33:02.456, Node N002, Modify, Stiffness 8200], [14:33:03.789, Insulation material, Ignore, Thermal conductivity 0.04]. The feature sequence is {[Type: 1, Thermal conductivity 0.04, Time: 01.123], [Type: 2, Stiffness 8200, Time: 02.456], [Type: 1, Thermal conductivity 0.04, Time: 03.789]}. Capture and extraction take approximately 0.1 seconds (10 operations). The sequence is stored as an in-memory list for comparison by S72 to ensure accurate reflection of subsequent user actions.

[0102] S72. By fusing the edit distance of the operation sequence, the consistency coefficient of parameter adjustment direction, and the penalty term for missing key cognitive nodes, the deviation between the incremental path and the expected cognitive path is quantified.

[0103] This paper quantifies the deviation between the incremental path and the expected cognitive path by fusing the edit distance of the operation sequence, the consistency coefficient of parameter adjustment direction, and the penalty for missing key cognitive nodes. It compares the incremental path of S71 with the expected path of S62 using multi-dimensional indicators to assess the degree of deviation of user behavior. The edit distance is defined as the minimum number of edit operations (insertion, deletion, replacement) in the operation sequence, calculated using the Levenshtein algorithm, ranging from [0, ∞]. The consistency coefficient is defined as the similarity of parameter adjustment directions, ranging from [0, 1], calculated using the cosine similarity of parameter vectors. The penalty for missing nodes is defined as the number of missing key nodes (such as modifying thermal conductivity) in the expected path, ranging from [0, ∞). The deviation is defined as a weighted sum of three terms: 0.5 × normalized edit distance + 0.3 × consistency coefficient + 0.2 × penalty for missing nodes. The calculation process includes: first, calculating the edit distance between the incremental path and the expected path (implemented in Python); second, comparing the cosine similarity of parameter vectors (calculated using NumPy); third, counting the number of missing key nodes (through node ID matching); and finally, fusing the three terms to obtain the deviation. The paths are obtained from S71 and S62, and the calculations are performed using Python libraries.

[0104] For example, in the scenario above, the S71 incremental path contains 3 operations, while the S62 expected path contains 10 operations (5 view, 3 modify, 2 verify). The edit distance is calculated as 7 using the Levenshtein algorithm (requiring 7 operations for alignment), normalized to 7 / 10 = 0.7. The consistency coefficient is based on thermal conductivity and stiffness vectors, with a cosine similarity of 0.85. The missing value penalty is 2 (for expected modifications to thermal conductivity and missing verification nodes). The deviation is calculated as 0.5 × 0.7 + 0.3 × 0.85 + 0.2 × 2 = 1.005. The calculation takes approximately 0.2 seconds (for 10 operations). The deviation is stored as a memory variable for S73 to determine, ensuring accurate quantification of path differences.

[0105] S73. When the deviation exceeds the preset deviation threshold or the system time reaches the latest cognitive time limit, the cognitive enhancement engine is activated. Based on the expected cognitive path, the cognitive enhancement engine generates a multi-dimensional cognitive enhancement plan and performs progressive cognitive reinforcement intervention on the user until the user generates a behavioral signal to actively accept intervention again or the path deviation falls back to within the safe threshold.

[0106] When the deviation exceeds the threshold or reaches the latest cognitive time limit, the cognitive enhancement engine is activated to generate a multi-dimensional cognitive enhancement plan for progressive intervention. This means that by detecting the deviation of S72 and the time limit of S63, the enhancement intervention is triggered, generating a plan that includes visual elements, prompts, and feedback to guide the user to accept the intervention. The deviation threshold is defined as 0.8, based on engineering experience. The cognitive enhancement engine is defined as a rule-based intervention generator. The input is the deviation and the expected path, and the output is the enhancement plan (highlight parameters, pop-up prompts, performance feedback). The multi-dimensional plan includes visual enhancement (RGB contrast [0, 1]), prompt frequency (times / minute), and feedback content (performance change description). The intervention process includes: First, comparing the deviation with 0.8 or checking if the 13.6-second time limit has been reached. If triggered, the engine generates a plan: highlight parameters (RGB [255, 0, 0]), pop-ups 2 times / minute, and feedback stiffness change. Progressive enhancement increases the contrast by 10% or provides 1 prompt per minute until the user modifies the parameters or views the information 3 times. Interventions are implemented via the Revit API, and scheme generation is performed using Python.

[0107] For example, in the scenario above, the S72 deviation of 1.005 > 0.8, triggering the cognitive enhancement engine. The engine, based on the S62 path generation scheme, displays the following: a high-brightness thermal conductivity of 0.042 (RGB[255, 0, 0]), a pop-up window twice per minute (prompting "Adjusting the thermal conductivity can improve stiffness"), and feedback that "Stiffness can be increased to 7600 kN·m". 2 The Revit API intervened, detecting a user modification of the thermal conductivity to 0.042 after 1 minute, and stopped the intervention. Generation and application took approximately 0.3 seconds. The solution was stored as a log to ensure user acceptance of the intervention.

[0108] The aforementioned technical solution constructs an operational influence propagation chain ending with a rejection behavior and calculates the contribution of operational nodes to accurately infer the historical spatiotemporal range of user rejection behavior. Then, based on historical interaction data, it uses a graph neural network to generate user decision-making pattern feature vectors and matches them with a pre-built knowledge base to identify target causes (such as insufficient cognition, time pressure, or lack of trust). Based on the cause type, it retrieves similar case cognitive trajectories and, combined with user profiles, uses a cognitive dynamics model to plan the expected cognitive path, including triggering events, deepening operations, and verification nodes, setting a latest cognitive time limit based on risk adjustment factors. By monitoring the deviation between the user's subsequent decision path and the expected path in real time, multi-dimensional cognitive enhancement intervention is triggered when the deviation exceeds the limit or the time limit is reached. The system can effectively guide users to re-cognize optimization potential, significantly improve acceptance of implicit heuristic interventions, reduce suboptimal decisions caused by human cognitive biases, improve the automation level and design efficiency of prefabricated building structural mechanical performance optimization, and ultimately ensure the safety and economy of building structures.

[0109] Figure 2This application provides a schematic diagram of a BIM-based prefabricated building structure mechanical performance optimization system, as shown in the embodiment. Figure 2 As shown, the system includes:

[0110] The capture and deconstruction module 100 is used to capture the user's operation sequence in real time and deconstruct its optimization decision path when the user optimizes the structural mechanical performance of prefabricated buildings based on the BIM model.

[0111] The comparison and identification positioning module 200 is used to compare with a pre-set optimization knowledge base based on the optimization decision path to identify and locate optimization blind spots that users may overlook; wherein, optimization blind spots include at least the coupling relationship between the performance of thermal insulation materials and the mechanical behavior of structural nodes;

[0112] The implicit heuristic intervention module 300 is used to provide implicit heuristic intervention to users based on optimization blind spots. Specifically, it reveals the optimization potential in the optimization blind spots through parameterized fine-tuning and adaptively intervenes based on user cognitive feedback until it guides users to complete the optimization autonomously.

[0113] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for optimizing the mechanical performance of prefabricated building structures based on BIM, characterized in that, include: When users optimize the structural mechanical performance of prefabricated buildings based on BIM models, the system captures the user's operation sequence in real time and deconstructs their optimization decision path. Based on the optimized decision path, it is compared with the pre-built optimization knowledge base to identify and locate optimization blind spots that users may overlook; among them, optimization blind spots include at least the coupling relationship between the performance of thermal insulation materials and the mechanical behavior of structural nodes; Based on the optimization blind spot, implicit heuristic intervention is carried out on users. Specifically, the optimization potential in the optimization blind spot is revealed through parameter fine-tuning, and the intervention process is adaptively adjusted according to user cognitive feedback until the user completes the optimization on their own.

2. The method for optimizing the mechanical performance of prefabricated building structures based on BIM as described in claim 1, characterized in that, The real-time capture of user operation sequences and deconstruction of their optimization decision paths include: Real-time monitoring and capture of user interaction events with the BIM model, forming a user operation sequence sorted by timestamp; wherein the user operation sequence includes at least: modification operations on component geometric properties, material parameters and node connection methods; Analyze the temporal relationships and logical connections between operations in the user's operation sequence to deconstruct the decision-making steps taken by the user to achieve specific mechanical performance goals; Based on the deconstructed decision steps, a structured optimized decision path is generated.

3. The method for optimizing the mechanical performance of prefabricated building structures based on BIM as described in claim 1, characterized in that, The step of comparing the optimized decision path with a pre-built optimization knowledge base to identify and locate optimization blind spots that users may overlook includes: Construct an abstract network representation of a pre-built optimized knowledge base; where nodes in the abstract network representation represent design entities and performance indicators, and edges represent the mechanical coupling and constraint relationships between them; The optimized decision path is mapped to an abstract network representation, and a semantic matching operation is performed to generate a path coverage graph and a relation-ignoring weight graph. The optimization decision path is based on path coverage graph analysis to improve the coverage of the network structure, and the optimization decision path is based on relation neglect weight graph analysis to improve the attention of specific mechanical coupling relationships. Network areas with coverage completeness below a preset coverage completeness threshold and attention below the corresponding attention threshold are identified as optimization blind spots.

4. The method for optimizing the mechanical performance of prefabricated building structures based on BIM as described in claim 1, characterized in that, The process of revealing optimization potential in the optimization blind spot through parameterized fine-tuning, and adaptively intervening based on user cognitive feedback until guiding the user to complete the optimization autonomously, includes: A global sensitivity analysis was conducted on the key design parameters corresponding to the optimization blind zone to identify the set of parameters that have a significant impact on the overall mechanical performance. Apply limited, reversible, automated adjustments to the parameter set in accordance with engineering specifications, and generate corresponding exploratory optimization schemes; Monitor user interaction with the tentative optimization plan in real time and update the user's cognitive state accordingly; When the updated user cognitive state confirms that the user is in a state of continuous neglect, the visibility of the exploratory optimization scheme is progressively increased until the user's adoption behavior or active interaction signal is captured.

5. The method for optimizing the mechanical performance of prefabricated building structures based on BIM as described in claim 1, characterized in that, Also includes: When a user rejects an implicit heuristic intervention, the historical spatiotemporal scope used to analyze the reasons for the rejection is inferred based on the context in which the user's rejection occurred. Based on historical interaction data recorded within a historical time and space range, we analyze the reasons why users reject implicit heuristic interventions. Based on the objective reasons, plan the expected cognitive path for users to re-understand the optimization potential, and set the latest time limit for understanding; Monitor the deviation between the user's subsequent optimization decision path and the expected cognitive path, and trigger cognitive enhancement intervention when the deviation exceeds the limit or the latest cognitive time limit is reached.

6. The method for optimizing the mechanical performance of prefabricated building structures based on BIM as described in claim 5, characterized in that, The inference of the historical spatiotemporal range for analyzing the reasons for rejection based on the context in which the user's rejection behavior occurred includes: Extract key operation timestamp sequences related to optimization blind spots from user operation sequences, and combine them with BIM model version iteration logs to construct an operation impact propagation chain ending with rejection behavior; Based on the semantic similarity and time decay weight of each operation node in the operation influence propagation chain, and by introducing the coupling strength factor of the structural mechanical parameter changes between nodes, the contribution of each operation node to the rejection behavior is calculated. The time interval and model space range covered by the set of operation nodes whose contribution exceeds the preset contribution threshold are defined as the historical spatiotemporal range.

7. The method for optimizing the mechanical performance of prefabricated building structures based on BIM as described in claim 5, characterized in that, The analysis of the target reasons that lead users to reject implicit heuristic interventions, based on interaction history data recorded within the historical spatiotemporal range, includes: Within a historical timeframe, extract user response latency, modification frequency, and parameter rollback behavior to exploratory optimization schemes, and use this data as historical interaction data. A user decision-making behavior graph is constructed based on historical interaction data, and a graph neural network is used to generate feature vectors of user decision-making patterns. The feature vector is matched with typical rejection pattern templates in a pre-built optimized knowledge base in multiple dimensions, and the target cause is identified based on the matching results; wherein the target cause includes at least one of the following: Insufficient understanding of the coupling relationship between insulation materials and structural nodes, simplified decision-making strategies adopted due to time pressure, and a lack of trust in automated adjustments based on historical experience.

8. The method for optimizing the mechanical performance of prefabricated building structures based on BIM as described in claim 5, characterized in that, Based on the stated objective reasons, the plan outlines the expected cognitive path for users to re-understand the optimization potential, and sets a latest cognitive deadline, including: Based on the type of the target cause, retrieve the cognitive evolution trajectory of similar cases from the optimized knowledge base and extract the sequence of key cognitive turning points; Based on the learning ability indicators and historical operation proficiency in user profile data, a cognitive dynamics model is used to simulate the rational cognitive process from a state of rejection to voluntary acceptance of intervention, generating a staged cognitive path that includes a sequence of cognitive triggering events, an expected cognitive deepening operation chain, and cognitive verification nodes, which serves as the expected cognitive path. Based on the estimated total time taken for each stage in the expected cognitive path, and by introducing a risk adjustment factor that is dynamically adjusted according to the critical impact of the optimization blind spot on the structural mechanical performance, the latest cognitive time limit is set.

9. The method for optimizing the mechanical performance of prefabricated building structures based on BIM as described in claim 5, characterized in that, The monitoring determines the deviation between the user's subsequent optimization decision path and the expected cognitive path, and triggers cognitive enhancement intervention when the deviation exceeds the limit or the latest cognitive time limit is reached, including: Real-time monitoring of the incremental optimization decision path after user rejection, and extraction of operational feature sequences; By integrating the edit distance of the operation sequence, the consistency coefficient of parameter adjustment direction, and the penalty term for missing key cognitive nodes, the deviation between the incremental path and the expected cognitive path is quantified. When the deviation exceeds the preset deviation threshold or the system time reaches the latest cognitive time limit, the cognitive enhancement engine is activated. Based on the expected cognitive path, the cognitive enhancement engine generates a multi-dimensional cognitive enhancement plan and performs progressive cognitive reinforcement intervention on the user until the user generates a behavioral signal to actively accept intervention again or the path deviation falls back to within the safe threshold.

10. A BIM-based system for optimizing the mechanical performance of prefabricated building structures, characterized in that, include: The capture and deconstruction module is used to capture the user's operation sequence in real time and deconstruct its optimization decision path when the user optimizes the structural mechanical performance of prefabricated buildings based on the BIM model. The comparison and identification module is used to compare the optimized decision path with a pre-set optimization knowledge base to identify and locate optimization blind spots that users may overlook; among them, optimization blind spots include at least the coupling relationship between the performance of thermal insulation materials and the mechanical behavior of structural nodes; The implicit heuristic intervention module is used to provide implicit heuristic intervention to users based on optimization blind spots. Specifically, it reveals the optimization potential in the optimization blind spots through parameterized fine-tuning and adaptively intervenes based on user cognitive feedback until it guides users to complete the optimization autonomously.

Images