BIM project progress management platform and method based on space-time association network

Abstract

The invention provides a BIM project progress management platform and method based on a space-time association network, and the method comprises the steps: building a component-process-resource dynamic topology through expanding an IFC standard, and carrying out the unified modeling of space coordinates, time dependence and resource weights; a key path risk is predicted by using a space-time diagram convolutional network, a feasible scheme is generated through Monte Carlo simulation, and industry specifications are embedded for automatic verification; a double-layer feedback mechanism is introduced, and model parameters and a scheduling strategy are continuously optimized through field data; and in cooperation with time-sharing grading LOD lightweight rendering, real-time interaction of twenty thousand-level components is realized. The platform significantly shortens the deviation response time, reduces the manual dependence and violation risk, and completes the crossing from static monitoring to dynamic self-optimization.

Description

Technical Field

[0001] This invention belongs to the field of information technology, specifically relating to a BIM project progress management platform and method based on spatiotemporal correlation networks. Background Technology

[0002] Existing BIM project progress management systems generally adopt a modular, layered architecture, with core processes encompassing five major modules: data management, progress tracking, visualization, resource scheduling, and compliance checks. In terms of data management, the system typically stores structured data such as design models and schedules in relational databases like MySQL, and extracts the geometric and attribute information of components using an IFC standard parser. Unstructured data, such as construction logs and supervision reports, are stored as PDFs or images on file servers, relying on manual periodic archiving. To achieve data association, existing solutions simply bind BIM components to Gantt chart task nodes via process IDs in the time dimension, and establish a static R-Tree spatial index based on the component's geometric coordinates in the spatial dimension. Once created, the index is not dynamically updated with construction progress.

[0003] In the progress tracking phase, the system first uses the Critical Path Method (CPM) to generate an initial schedule, and the project manager manually enters resource capacity constraints such as tower cranes and manpower. Then, on-site personnel manually report actual progress daily (e.g., concrete pouring completion rate), and database triggers update the Gantt chart status. When the deviation exceeds a preset threshold (e.g., Δ>10%), the system pushes an alarm message to the project manager via email. The visualization module uses the Three.js engine to provide a lightweight display of all IFC component geometric data on the web, supporting component filtering and attribute viewing by floor or specialty. Delayed progress areas are marked on the 3D model surface using red, yellow, and green colors, with the color intensity determined solely by the absolute value of the difference between the planned and actual time.

[0004] Resource scheduling relies entirely on manual decision-making: after receiving a delay report, the project manager needs to consult historical experience in the Excel ledger and then manually adjust the work sequence or add more work teams; resource conflict detection is completed manually by comparing the Gantt chart resource load curve with the equipment ledger, lacking an automated early warning mechanism. Regarding compliance checks, the existing system converts clauses from the "Construction Project Management Standard" and other regulations into executable SQL conditions (e.g., "Daily concrete pouring volume ≤ 2000m³"). 3 The rule base version is maintained by a dedicated person on a regular basis, and verification queries are executed in batches. It cannot be updated in real time with changes in the field.

[0005] In summary, although the existing system can achieve basic progress management, it has exposed three major technical defects in practical applications: First, time and space data are stored independently, lacking component-level 4D spatiotemporal topology relationships, which makes it impossible to detect and correct dynamic conflicts between the steel column installation coordinates and subsequent concrete pouring procedures in a timely manner; Second, the rule base is static and relies on manual maintenance, making it impossible to integrate unstructured data such as construction logs and supervisor handwritten records in real time, resulting in progress adjustments lagging behind the actual situation on site; Third, conventional WebGL-based rendering engines (such as Three.js) suffer from an interaction delay of more than 3 seconds when loading more than 20,000 IFC components due to the lack of a time-sharing and hierarchical LOD loading strategy, making it difficult to meet the real-time decision-making needs on site. Summary of the Invention

[0006] In view of this, the present invention provides a BIM project schedule management platform and method based on spatiotemporal correlation network. By extending the spatiotemporal attributes of components defined by the IFC standard, a dynamic topology network of components-processes-resources is constructed; the risk of critical path breakage is predicted by using spatiotemporal graph convolutional network; the self-evolution of model parameters and scheduling strategies is realized through a two-layer feedback mechanism; and lightweight interaction of ultra-large-scale models is realized by using time-division hierarchical level of detail (LOD) technology.

[0007] The technical solution for implementing the present invention is as follows:

[0008] A BIM project schedule self-optimization management platform based on spatiotemporal correlation networks includes:

[0009] The data layer is used to integrate multi-source data and build a spatiotemporal data warehouse. The multi-source data includes structured IFC models, unstructured construction logs, and real-time sensor data. The spatiotemporal data warehouse defines spatiotemporal topological relationship entities by extending the IFC4x3 standard, and stores the three-dimensional spatial coordinates of components, process timelines, resource associations, and compliance rule base.

[0010] The model layer is used to construct a dynamic topology graph of components, processes, and resources based on a spatiotemporal correlation network. The nodes contain spatial coordinates, temporal dependencies, and resource weights, and the deviation between the actual data and the planning model is calculated in real time through a dynamic mapping engine.

[0011] The algorithm layer is used to deploy a spatiotemporal graph convolutional network to predict the risk of critical path breakage. It combines Monte Carlo simulation to generate multiple alternative candidates and achieves model parameter self-evolution and scheduling strategy optimization through a two-layer feedback system.

[0012] The application layer is used to automatically generate schedule adjustment plans based on the results of intelligent simulation, synchronously update material requirements plans and labor lists, and verify compliance in real time.

[0013] The interaction layer is used to achieve lightweight visualization through a time-sharing and hierarchical LOD loading algorithm.

[0014] Furthermore, the spatiotemporal topological relation entity includes:

[0015] Unique identifier for components;

[0016] A 3D bounding box is used to store component coordinates and perform spatial collision detection.

[0017] Process dependency chain, which records the set of preceding process IDs to build time-dimensional dependencies;

[0018] Resource impact factors quantify resource utilization priorities.

[0019] Furthermore, the dynamic adjacency matrix generation algorithm includes:

[0020] Time conflict weight: 1 / (1+e^(-Δt)), where Δt is the actual delay duration;

[0021] Spatial conflict weight: Spatial overlap rate × Resource coupling coefficient.

[0022] Furthermore, the two-layer feedback system includes:

[0023] Data layer feedback: The weight parameters of the GCN model are adjusted in real time based on construction acceptance data;

[0024] Feedback from the decision-making level: The reinforcement learning module based on the deep Q-network has a reward function of R = 1 - (actual delay time / predicted delay time).

[0025] Furthermore, the time-sharing hierarchical LOD loading algorithm includes:

[0026] Spatial grading: When the line of sight is <50 meters, full geometric accuracy is applied; when the line of sight is ≥50 meters, it is simplified to an axis-aligned bounding box.

[0027] Time-based grading: Non-critical processes are delayed in loading, while critical processes are rendered first.

[0028] A method for dynamically predicting the progress of a BIM project based on the aforementioned platform includes the following steps:

[0029] S1: Construct spatiotemporal topological relationship entities by extending the IFC4x3 standard, and associate components with processes;

[0030] S2: A dual-channel Transformer model is used to process unstructured data, extract time-sensitive entities, and map them to spatiotemporal network nodes;

[0031] S3: Predict critical path breakage risk based on spatiotemporal graph convolutional network and generate multiple alternative candidates;

[0032] S4: Automatically verify the legality of the solution by embedding compliance constraints through Monte Carlo simulation;

[0033] S5: Optimize scheduling strategies based on reinforcement learning, output adjustment plans, and update the schedule.

[0034] Beneficial effects:

[0035] (1) Deep integration of spatiotemporal correlation improves dynamic correction efficiency.

[0036] This invention extends the IFC standard definition of spatiotemporal topological relationship entities to construct a dynamic network model containing 12 core attributes (3D coordinates, process dependency chains, resource weights, etc.), achieving deep coupling of spatiotemporal data. Implementation verification shows that the schedule deviation response time is reduced from 8 hours in the traditional system to 3.2 hours, and the spatiotemporal topological relationship coverage is increased to 98.7% (compared to 76.2% in the traditional solution), effectively solving the decision-making lag problem caused by spatiotemporal decoupling.

[0037] (2) Intelligent algorithms drive decision-making, significantly reducing reliance on human intervention.

[0038] This invention introduces a Spatiotemporal Graph Convolutional Network (ST-GCN) and Monte Carlo simulation. The input includes node features containing time difference (Δt), resource utilization rate (η), and a dynamic adjacency matrix. The output is the critical path breakage probability for the next 72 hours (accuracy 93.5%), and it automatically generates multiple candidate solutions (such as parallel processing and resource allocation). Through a dual-layer feedback mechanism (data correction + reinforcement learning), the system can self-optimize its scheduling strategy, reducing tower crane idle time by 18% and improving resource scheduling efficiency by over 40%, achieving a leap from "human decision-making" to "data intelligence."

[0039] (3) Breakthrough in lightweight interactive technology, efficient rendering of ultra-large-scale models

[0040] This invention develops a time-sharing and hierarchical LOD loading algorithm. Through spatial hierarchical (dynamic precision adjustment of view distance) and temporal hierarchical (prioritizing loading of critical processes), it reduces memory usage to 1.8GB (industry average 3.5GB) and first-screen rendering time to <1.2 seconds. It supports real-time component-level sectioning, measurement, and annotation (response latency ≤0.5 seconds) and automatically generates multi-format reports including carbon emissions and resource idle rates (updated every 6 hours). Its interactive performance and data visualization capabilities are significantly superior to existing solutions.

[0041] (4) Automated compliance verification enables more precise risk control.

[0042] This invention embeds dynamic compliance constraint functions (such as check_night_work) into Monte Carlo simulations, automatically compares and adjusts the scheme against standards such as GB / T 51301-2018, and outputs a risk list in real time. Compared with manual verification, compliance verification efficiency is improved by 60%, and the risk omission rate is reduced from 25% to <5%, significantly reducing the risk of construction violations. Attached Figure Description

[0043] Figure 1 This is a system architecture diagram of the present invention.

[0044] Figure 2 This is a schematic diagram of the spatiotemporal correlation network node structure.

[0045] Figure 3 This is a diagram illustrating the lightweight LOD loading effect. Detailed Implementation

[0046] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0047] This invention proposes a BIM project schedule self-optimization management platform based on spatiotemporal correlation networks, such as... Figure 1 As shown, the overall architecture adopts a five-layer structure, namely the data layer, model layer, algorithm layer, application layer, and interaction layer. Through deep fusion of spatiotemporal data, intelligent algorithm inference, and lightweight visualization technology, dynamic self-optimization of project progress is achieved.

[0048] The data layer is responsible for aggregating multi-source heterogeneous information throughout the entire project lifecycle: structured data includes design models and schedules conforming to the IFC4x3 standard; unstructured data includes construction logs, supervision reports, and site photos; and real-time sensor data covers indicators such as tower crane operation, concrete pouring volume, and environmental monitoring. The platform extends the IFC4x3 standard by adding a "spatiotemporal topology entity (IfcRelSpaceTime)," assigning each component a unique global identifier and associating it with a 3D spatial bounding box (x_min, y_min, z_min, x_max, y_max, z_max), a process dependency chain (a set of preceding process IDs), and resource impact factors (e.g., tower crane weight 0.7, concrete pump truck weight 0.3). The spatiotemporal data warehouse constructed in this way can simultaneously store the spatial coordinates of components, timeline information, resource relationships, and an industry compliance rule base, providing a unified and scalable data foundation for subsequent analysis.

[0049] The model layer establishes a dynamic topology graph of "component-process-resource" based on the above data. Each node comprehensively characterizes multi-dimensional features such as spatial location, temporal dependence, and resource weight; the weights of the edges between nodes are calculated in real time by a dynamic mapping engine. This engine, based on the Apache Kafka stream processing framework, continuously captures on-site data, calculates the deviation Δ between actual and planned time (Δ = actual time / planned time - 1), and maps the result to the network edges, forming a real-time updated spatiotemporal relational network, such as... Figure 2 As shown.

[0050] The algorithm layer deploys two core algorithms:

[0051] Spatiotemporal Graph Convolutional Network (ST-GCN). Node feature vectors are composed of spatial coordinates (x, y, z), the planned and actual time difference Δt, and resource occupancy rate η. The adjacency matrix is ​​dynamically generated based on "temporal conflict weight" and "spatial collision weight"—if the completion time of component A's process is later than the start time of component B's process, the temporal conflict weight is 1 / (1+e^(-Δt)); if two components spatially overlap and occupy the same type of resources, the spatial collision weight is the spatial overlap rate multiplied by the resource coupling coefficient; otherwise, the weight is reset to zero. After this adjacency matrix is ​​input into the GCN, it can output the probability of critical path failure in the next 72 hours. When the confidence level is ≥90%, the platform highlights the risk area affecting ≥3 subsequent processes in red on the 3D heatmap.

[0052] Monte Carlo simulation and compliance constraint injection. The platform randomly samples and generates a large number of schedule adjustment schemes in the simulation space, and calls a compliance verification function in each iteration, such as checking whether the nighttime construction time exceeds 72 hours. If it violates Article 5.2.3 of the "Regulations on the Management of Construction Environment", the scheme is immediately eliminated; otherwise, its comprehensive suitability (comprehensive schedule, cost, and resource utilization) is calculated, and several feasible candidate schemes are finally output for decision-making.

[0053] The application layer automatically generates schedule adjustment plans based on the algorithm layer results, including process rescheduling, resource reallocation, and synchronous updates of Material Requirements Planning (MRP) and the bill of materials. A built-in compliance verification module compares the plans with industry standards such as GB / T 51301-2018 in real time, outputting a risk list and rectification suggestions. The entire process requires no manual intervention, achieving a closed loop of "data → algorithm → decision → execution".

[0054] The interaction layer features a lightweight visualization engine that supports real-time sectioning, annotation, and measurement of over 20,000 components, with a response latency of ≤0.5 seconds. Figure 3 As shown, the engine employs a time-sharing, hierarchical LOD strategy:

[0055] Spatial grading—when the viewing distance is <50m, the complete geometry is loaded (triangular facet accuracy ≤0.5mm), and when the viewing distance is ≥50m, it is simplified to axis-aligned bounding box (AABB), reducing memory usage by 48%;

[0056] Time-based grading – dynamically adjusts loading priority based on the criticality of the process, delays loading of non-critical path models, and ensures first-screen rendering time is less than 1.2 seconds.

[0057] In addition, the platform automatically generates intelligent reports in PDF / Excel format every 6 hours, covering indicators such as carbon emissions and resource idle rate, to assist multi-departmental collaborative decision-making.

[0058] To ensure the algorithm continuously aligns with the real-world environment, the platform introduces a "two-layer feedback self-evolution mechanism":

[0059] Data layer feedback – After every 100 process nodes are completed, the weights of the GCN model are fine-tuned online using actual project duration and resource consumption data to improve the model’s adaptability to changes on site.

[0060] Decision-making feedback – a reinforcement learning module based on deep Q-network (DQN) with the reward function R = 1 - (actual delay time / predicted delay time) as the optimization objective. Through 30 days of continuous training, the prediction error rate is reduced to below 5%, achieving self-optimization of the scheduling strategy.

[0061] Through the above technologies and mechanisms, this invention upgrades the traditional "static monitoring" to "dynamic self-optimization", significantly shortens the response time of schedule deviations, reduces reliance on manual intervention, improves the interaction efficiency of ultra-large-scale models, and forms a self-evolving engineering schedule management system.

[0062] Implementation verification

[0063] This platform achieves:

[0064] Data correlation: The spatiotemporal topology coverage (98.7%) directly reflects the effectiveness of IFC standard extension and multi-source data fusion, and compared with the traditional solution (76.2%), it demonstrates the solution to the spatiotemporal decoupling problem;

[0065] Algorithm effectiveness: The comparison of the critical path break prediction accuracy (93.5%) with the GCN benchmark model (81.2%) demonstrates the superiority of the dynamic adjacency matrix algorithm;

[0066] Engineering value: Tower crane idle time is reduced by 18%, and the response time for schedule deviations is reduced to 3.2 hours. This quantitatively demonstrates the efficiency improvement brought about by intelligent scheduling and lightweight interaction, forming a technology-effect closed loop with the "self-evolution management mechanism".

[0067] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A BIM project schedule self-optimization management platform based on spatiotemporal correlation networks, characterized in that, include: The data layer is used to integrate multi-source data and build a spatiotemporal data warehouse. The multi-source data includes structured IFC models, unstructured construction logs, and real-time sensor data. The spatiotemporal data warehouse defines spatiotemporal topological relationship entities by extending the IFC4x3 standard, and stores the three-dimensional spatial coordinates of components, process timelines, resource associations, and compliance rule base. The model layer is used to construct a dynamic topology graph of components, processes, and resources based on a spatiotemporal correlation network. The nodes contain spatial coordinates, temporal dependencies, and resource weights, and the deviation between the actual data and the planning model is calculated in real time through a dynamic mapping engine. The algorithm layer is used to deploy a spatiotemporal graph convolutional network to predict the risk of critical path breakage. It combines Monte Carlo simulation to generate multiple alternative candidates and achieves model parameter self-evolution and scheduling strategy optimization through a two-layer feedback system. The application layer is used to automatically generate schedule adjustment plans based on the results of intelligent simulation, synchronously update material requirements plans and labor lists, and verify compliance in real time. The interaction layer is used to achieve lightweight visualization through a time-sharing and hierarchical LOD loading algorithm.

2. The management platform as described in claim 1, characterized in that, The spatiotemporal topological relationship entity includes: Unique identifier for components; A 3D bounding box is used to store component coordinates and perform spatial collision detection. Process dependency chain, which records the set of preceding process IDs to build time-dimensional dependencies; Resource impact factors quantify resource utilization priorities.

3. The platform according to claim 2, characterized in that, The dynamic adjacency matrix generation algorithm includes: Time conflict weight: 1 / (1+e^(-Δt)), where Δt is the actual delay duration; Spatial conflict weight: Spatial overlap rate × Resource coupling coefficient.

4. The platform according to claim 1, characterized in that, The two-layer feedback system includes: Data layer feedback: The weight parameters of the GCN model are adjusted in real time based on construction acceptance data; Feedback from the decision-making level: The reinforcement learning module based on the deep Q-network has a reward function of R = 1 - (actual delay time / predicted delay time).

5. The platform according to any one of claims 1-4, characterized in that, The time-sharing and hierarchical LOD loading algorithm includes: Spatial grading: When the line of sight is <50 meters, full geometric accuracy is applied; when the line of sight is ≥50 meters, it is simplified to an axis-aligned bounding box. Time-based grading: Non-critical processes are delayed in loading, while critical processes are rendered first.

6. A method for dynamic simulation of BIM project progress based on the platform described in claim 1, characterized in that, Includes the following steps: S1: Construct spatiotemporal topological relationship entities by extending the IFC4x3 standard, and associate components with processes; S2: A dual-channel Transformer model is used to process unstructured data, extract time-sensitive entities, and map them to spatiotemporal network nodes; S3: Predict critical path breakage risk based on spatiotemporal graph convolutional network and generate multiple alternative candidates; S4: Automatically verify the legality of the solution by embedding compliance constraints through Monte Carlo simulation; S5: Optimize scheduling strategies based on reinforcement learning, output adjustment plans, and update the schedule.

Images