Intelligent perception and early warning system and method for long-span space structure based on BIM and improved BP neural network

Abstract

This invention discloses an intelligent sensing and early warning system and method for large-span spatial structures based on BIM and an improved BP neural network. The system includes a physical sensing layer, a data integration and twin layer, an intelligent analysis and early warning layer, and an interactive application layer. The physical sensing layer collects multi-dimensional physical quantities of the structure. The data integration and twin layer realizes bidirectional conversion between BIM and FEA models through a BIM-FEA integration engine and constructs a dynamic digital twin model in conjunction with a monitoring data access gateway. The intelligent analysis and early warning layer realizes structural response prediction and anomaly identification and generates graded early warnings through an improved physical-guided BP neural network. The interactive application layer realizes three-dimensional visualization of early warning information and cross-regional collaborative management through a visual early warning plugin and a cloud-based collaborative management platform. The method includes steps such as constructing a digital twin model base, dynamic semantic fusion of multi-source data, BP neural network prediction and anomaly detection, graded early warning visualization and release, and model feedback iterative optimization.

Description

Technical Field

[0001] This invention belongs to the field of civil engineering structural health monitoring and intelligent operation and maintenance technology. Specifically, it relates to an intelligent sensing and early warning system and method for large-span spatial structures based on Building Information Modeling (BIM) and improved BP neural networks. In particular, it relates to a system and method for intelligent sensing, visualization and early warning of the entire life cycle of large-span spatial structures that integrates the Internet of Things (IoT), finite element analysis (FEA) and physically guided BP neural networks. Background Technology

[0002] Existing large-span spatial structures, such as national stadiums, large airport terminals, and convention centers, are core symbols of modern cities and major national infrastructure. Their structural safety is directly related to public safety, economic operation, and social stability. These structures are characterized by large spans, complex stresses, and harsh service environments. During long-term use, they are subjected to the coupled effects of multiple factors, including wind loads, temperature effects, material aging, fatigue, and extreme events. Performance degradation and damage evolution exhibit significant uncertainty and insidiousness. Once progressive damage accumulation or sudden failure occurs, it will cause catastrophic consequences. Therefore, achieving real-time, accurate perception, dynamic assessment, and early warning of the service status of large-span structures is an urgent need to ensure the safe operation of major projects.

[0003] Traditional structural health monitoring systems suffer from several limitations. Firstly, data fragmentation and insufficient visualization: monitoring data is independent of design information, often presented in two-dimensional charts, lacking an intuitive and dynamic connection to the three-dimensional physical structure, making risk identification difficult. Secondly, weak model interoperability and dynamic integration capabilities: there is no efficient and automated data flow channel between the BIM model in the design phase and the monitoring data in the operation and maintenance phase. BIM models are mostly static design information carriers, failing to dynamically integrate real-time monitoring data and mechanical analysis results, limiting their in-depth application in full lifecycle safety management. Thirdly, limited intelligence in early warning systems: existing early warning systems rely heavily on single threshold judgments or simple statistical analysis, struggling to handle the high-dimensional, strong temporal sequence, and nonlinear characteristics of monitoring data for large-span structures. They lack sensitivity and accuracy in identifying early, minor damage and complex damage patterns, and the early warning results lack interpretable correlation with the structural physical mechanisms. Fourthly, a lack of system integration and standardization: the technologies for sensing, transmission, processing, and early warning vary, lacking a unified standardized framework and integration platform, creating information silos, resulting in high system deployment costs, poor scalability, and difficulty in reusing results.

[0004] In recent years, digital technologies centered on BIM, IoT technologies represented by wireless sensor networks and smart sensors, and artificial intelligence technologies represented by machine learning have offered new possibilities for solving the aforementioned problems. Existing research has attempted to combine BIM with IoT for structural monitoring or apply neural networks for data analysis, but key bottlenecks remain: a deep integration and dynamic update mechanism between BIM and monitoring data and mechanical models has not yet been established; data-driven models are heavily reliant on massive amounts of labeled data and often lack integration with the physical laws of the structure, resulting in limited generalization ability and reliability in large-span structure scenarios where measured data is scarce; and a standardized, scalable, and systematic solution integrating "multi-source sensing, dynamic twinning, intelligent analysis, and interactive early warning" is lacking. Summary of the Invention

[0005] To address the shortcomings of existing large-span spatial structure health monitoring systems in terms of data-model fusion depth, early warning intelligence level, and system integration, this invention provides an intelligent sensing and early warning system and method for large-span spatial structures based on BIM and an improved BP neural network. This system achieves dynamic semantic fusion of BIM models and multi-source monitoring data, physically guided intelligent early warning algorithms, and cloud-based collaborative visualization interaction, significantly improving the real-time performance, accuracy, intuitiveness, and decision-making capability of large-span structure safety status assessment.

[0006] This invention is achieved through the following technical solution: Firstly, it provides an intelligent sensing and early warning system for large-span spatial structures based on BIM and an improved BP neural network, comprising a physical sensing layer, a data integration and twin layer, an intelligent analysis and early warning layer, and an interactive application layer. The physical sensing layer consists of various intelligent sensor nodes deployed at key locations of the large-span structure, including but not limited to fiber optic strain sensors, wireless accelerometers, laser displacement sensors, inclinometers, and temperature and humidity sensors, used to collect multi-dimensional physical quantities such as strain, acceleration, displacement, rotation angle, and environmental parameters of the structure in real time. The data integration and twin layer includes a BIM-FEA integration engine, a monitoring data access gateway, and a dynamic digital twin model. The BIM-FEA integration engine, based on Industrial Foundation Class (IFC) standards or dedicated data interfaces, enables automatic conversion of geometric, material, and boundary condition information from the BIM model to the finite element analysis model during the design phase, as well as the reverse mapping of FEA analysis results (stress, strain, displacement cloud maps) to the BIM model. The monitoring data access gateway is responsible for receiving, parsing, and preprocessing multi-source heterogeneous monitoring data streams from the physical sensing layer. The dynamic digital twin model serves as the core carrier, achieving real-time, dynamic, and semantic association and integrated storage of monitoring data, BIM geometric attributes, and FEA mechanical states through unique component identifiers (such as GUIDs). The intelligent analysis and early warning layer is the core computing unit, incorporating an improved physical-guided BP neural network prediction module, a damage identification module, and an early warning rule engine. This layer receives time-series monitoring data from the data integration layer and corresponding FEA simulation baseline data, utilizes the improved BP neural network for structural response prediction and anomaly pattern recognition, and combines preset multi-level early warning thresholds and expert rules to generate tiered early warning information. The interactive application layer includes a visual early warning plugin developed based on BIM platforms such as Revit, and a cloud-based collaborative management platform. This layer is responsible for the 3D visualization and interactive display of the dynamic digital twin model, real-time monitoring data, and analysis and early warning results, providing managers with an immersive interface for structural status monitoring, risk identification, and decision support.

[0007] Furthermore, the visualization early warning plugin is based on the Autodesk Revit API and implemented in C#. Its main functional modules include a measurement point layout module, a measurement point search module, a real-time data module, and a historical data module. The cloud-based collaborative management platform enables lightweight loading of BIM models and web-based 3D rendering. The backend adopts a microservice architecture, and its core functions include statistical dashboards, 3D interaction, early warning handling, and data analysis.

[0008] On the other hand, this invention provides a method for intelligent perception and early warning of large-span spatial structures based on BIM and an improved BP neural network, applied to the aforementioned system, comprising the following steps:

[0009] S1. Construct a dynamic digital twin model base for large-span structures. Based on design drawings and as-built data, establish a high-precision BIM model; through the BIM-FEA integration engine, generate a parametric finite element benchmark model corresponding to the physical structure; add scalable attribute sets to key monitoring components in the BIM model to connect monitoring data and mechanical state information, forming the initial digital twin model base.

[0010] The BIM-FEA integration engine works as follows:

[0011] S1.1 Forward Conversion (BIM->FEA): Parses the geometric entities, material properties and load case information in the BIM model, and automatically converts them into input files suitable for general finite element software based on the IFC general standard, while preserving the mapping relationship between component IDs and BIM models;

[0012] S1.2 Reverse Mapping (FEA->BIM): Read the result file output by the finite element software, extract the stress, strain, displacement, mode shape and other result data of each element / node, aggregate the result data to the corresponding BIM component according to the mapping relationship, and store it in the extended attribute set of the BIM model in the form of color gradient cloud map, numerical label or custom attribute.

[0013] S2. Dynamic semantic fusion of multi-source monitoring data and BIM-FEA model. During the construction or operation and maintenance phase, the sensor network is optimized based on structural mechanics analysis. Through the monitoring data access gateway, the real-time collected sensor data streams are automatically matched and associated with the pre-set component IDs in the BIM model based on their spatial location information. Simultaneously, the updated monitoring data drives rapid simulation or model correction of the finite element baseline model, and the updated simulation results are back-mapped to the corresponding component attributes in the BIM model, achieving dynamic, bidirectional fusion and unified expression of physical monitoring data, virtual BIM model, and FEA analysis results.

[0014] The specific implementation of dynamic semantic fusion between monitoring data and BIM models includes:

[0015] S2.1 Sensor Metadata Registration: Create a corresponding "sensor family" instance for each physical sensor in the BIM model and place it precisely at the corresponding monitoring point in the three-dimensional space. The family instance contains metadata attributes such as sensor ID, type, range, installation direction, and the ID of the component to which it belongs.

[0016] S2.2 Real-time data flow association: Locate the corresponding "sensor family" instance and its attached parent component in the BIM model through the sensor ID, write the monitoring value into the dynamic attribute field of the parent component and add a timestamp;

[0017] S2.3 Data-driven FEA rapid simulation: Using real-time monitored environmental parameters as input, it triggers a preset parametric finite element script to perform rapid simulation, obtains the theoretical structural response under the current environment, and executes a reverse mapping process to update the theoretical state cloud map in the BIM model.

[0018] S3. Structural Response Prediction and Anomaly Detection Based on Temporal Backpropagation Neural Networks. An improved three-layer backpropagation neural network temporal prediction model is constructed, specifically designed to predict sensor measurements at the next time step based on historical monitoring data. This model overcomes the limitations of traditional backpropagation neural networks that directly process static pattern recognition. By introducing a temporal window mechanism and a hybrid loss function, it can effectively capture the time dependencies and nonlinear dynamic features in the monitoring data.

[0019] The prediction formula for this model is:

[0020] ,

[0021] From input layer to output layer:

[0022]

[0023] Hidden layer activation:

[0024]

[0025] Hidden layer to output layer:

[0026]

[0027] Output layer activation:

[0028]

[0029] In the formula: Let i be the i-th input feature value; The connection weights between the i-th node in the input layer and the j-th node in the hidden layer are given. This is the bias term for the j-th node in the hidden layer; The weighted sum of the inputs to the j-th node in the hidden layer; Here is the activation function for the hidden layer; This is the activation output value of the j-th node in the hidden layer; The connection weight between the j-th node in the hidden layer and the node in the output layer; For output layer node bias terms; This is the weighted sum of the inputs for the output layer nodes; The activation function for the output layer; This represents the predicted output value of the neural network.

[0030] The training formula for the improved three-layer BP neural network time series prediction model includes:

[0031] Output layer weight update:

[0032]

[0033]

[0034] Hidden layer weight update:

[0035]

[0036]

[0037] Output layer error:

[0038]

[0039] Hidden layer error

[0040]

[0041] In the formula: This is the true target value; This represents the predicted output value of the neural network. For output layer error sensitivity; Let be the error sensitivity of the j-th node in the hidden layer; This is the learning rate.

[0042] S4. Tiered Early Warning Generation and BIM Visualization Integration and Release. Early warning rules integrate damage probability output from the BP neural network prediction module, key indicator threshold exceedances, and multi-sensor data correlation analysis results to generate multi-level early warning signals ranging from "Caution" to "Severe" according to preset rules. A visualization-based early warning plugin, developed using the Revit API, automatically binds early warning signals to associated components in the BIM model, providing 3D visual warnings through component color changes, flashing, and pop-up windows. Simultaneously, early warning information, associated monitoring data curves, and FEA cloud map comparisons are pushed to a cloud-based collaborative management platform, supporting real-time viewing and collaborative decision-making across multiple terminals and regions.

[0043] S5. Model Update and System Iterative Optimization Based on Early Warning Feedback. The system records the processing of each early warning event and the subsequent structural inspection / repair results, forming a case library. Using this feedback information, the BP model can be incrementally optimized, the parameters of the finite element model can be corrected, and the early warning threshold can be adaptively adjusted, thereby achieving continuous evolution of system performance.

[0044] The beneficial effects of this invention are: it achieves deep dynamic integration of data throughout the entire life cycle: through the innovative BIM-FEA integration engine and dynamic semantic mapping mechanism, this invention connects the three core data sources of design BIM, real-time monitoring, and mechanical simulation, and constructs a dynamic digital twin that can reflect the integrated "form-state-property" of the physical structure in real time. This solves the problem of severe separation between data and models in traditional monitoring and lays a high-fidelity data foundation for accurate perception and evaluation.

[0045] This invention enhances the intelligence and physical reliability of early warning algorithms by proposing an improved BP neural network that creatively integrates data-driven methods with the physical laws of structural mechanics. Through feature fusion or loss constraints, the learning process of the neural network is guided by physical knowledge, significantly improving the model's prediction accuracy, generalization ability, and interpretability of early warning results in scenarios with limited training data and large-span structures. This overcomes the shortcomings of purely data-driven models, which are often "black boxes" and prone to overfitting.

[0046] An intuitive and efficient 3D visualization interactive early warning system has been constructed: By developing a visualization early warning plugin and a web-based cloud collaboration platform deeply integrated with the BIM platform, abstract monitoring data, complex analysis results, and tiered early warning signals are mapped onto the real building information model in the most intuitive form, such as 3D color cloud maps, dynamic highlighting, and multi-dimensional information panels. This enables rapid risk location and understanding with a "what you see is what you get" approach, greatly improving the decision-making efficiency and collaborative handling capabilities of management personnel.

[0047] This invention has resulted in a standardized and scalable system-level solution: providing a complete system architecture and methodology from the perception layer, data layer, analysis layer to the application layer. Its data exchange based on open standards such as IFC, modular microservice design, and cloud-based collaborative deployment ensure the system's openness, integrability, and scalability. It can adapt to the monitoring needs of large-span spatial structures of different scales and types, facilitating the standardization and large-scale application of the technology. Attached Figure Description

[0048] Figure 1 : A schematic diagram of the overall architecture of the system described in this invention.

[0049] Figure 2 : Overall flowchart of the method described in this invention.

[0050] Figure 3 : Schematic diagram of the working principle of the BIM-FEA integrated engine.

[0051] Figure 4 A schematic diagram illustrating the dynamic integration of monitoring data and BIM model.

[0052] Figure 5: Schematic diagram of BP neural network structure.

[0053] Figure 6 A schematic diagram of the Revit-based visual alert plugin interface.

[0054] Figure 7 A schematic diagram of the functional interface of the cloud-based collaborative management platform. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the following embodiments are only for explaining the invention and are not intended to limit the scope of protection of the invention.

[0056] Example:

[0057] This embodiment uses the operational health monitoring and early warning of a large reticulated coal shed as an application scenario to implement the system and method described in this invention.

[0058] Step 1: Construct a high-precision BIM model of the large reticulated shell coal shed using architectural design drawings and as-built data. Based on the IFC standard, extract core data such as component type, component specifications, and load information from the BIM model. Complete finite element analysis using the BIM-FEA integrated module of this invention to obtain FEA analysis results such as stress and deformation cloud diagrams of the structure. Map these results back to the corresponding component attributes in the BIM model to form the initial digital twin of the coal shed structure.

[0059] Step 2: Based on the finite element analysis results, deploy a sensor network consisting of high-precision vibrating wire sensors in key areas such as high-stress zones and critical load-bearing members; send the real-time monitoring data and sensor IDs collected by the sensors to the project cloud platform server through an IoT gateway; retrieve the sensor list through Revit API and cloud platform API, create sensor families in the BIM model, and establish a precise mapping relationship between the sensors and corresponding structural members to achieve dynamic semantic fusion of monitoring data and the BIM-FEA model.

[0060] Step 3: Train and apply the BP neural network prediction model using historical sensor data. Select 14 days of data after the structure has been running stably, and input the time-series monitoring data and the corresponding FEA simulation benchmark data into the improved three-layer BP neural network time-series prediction model for training and verification. After the model's prediction accuracy meets the requirements for engineering applications, deploy the trained model to the intelligent analysis and early warning layer for real-time structural response prediction and anomaly detection.

[0061] Step 4: The intelligent analysis and early warning layer receives sensor monitoring data in real time. It then uses a trained BP neural network model to predict structural response and identify abnormal patterns. When the early warning module receives an anomaly identification signal, it combines the exceeding of key indicator thresholds and the correlation analysis results of multi-sensor data to determine the hazard level according to preset rules, generating multi-level early warning signals from "Caution" to "Severe". In the Revit visualization early warning plugin, abnormal components are highlighted and flashed, and core monitoring data such as the component's real-time strain curve are displayed. Simultaneously, alarm information is published on the cloud-based collaborative management platform and pushed to the safety manager's mobile terminal via SMS, WeChat, and other means.

[0062] Step 5: For the issued "Severe" level warning signal, arrange professional technicians to conduct on-site verification and testing of abnormal components using high-precision instruments. Based on the test results, formulate and implement corresponding repair and reinforcement treatment plans. The system records the entire process of handling the warning event, test data, repair results, and other information and incorporates them into the case library. This feedback information is used to incrementally learn and optimize the BP neural network model, correct the parameters of the finite element model, and adaptively adjust the warning threshold to achieve continuous evolution of system performance.

[0063] This embodiment achieves intelligent perception and early warning of the entire life cycle of a large reticulated coal shed through the above steps, solving the problems of data fragmentation, low early warning accuracy, and poor visualization in traditional monitoring systems. It effectively improves the safety management level of the coal shed structure during its operation and verifies the feasibility and effectiveness of the system and method of this invention in practical engineering applications.

[0064] It should be understood that the above embodiments are merely exemplary implementations of the present invention. The technical solutions of the present invention are not only applicable to coal sheds with large reticulated shell structures, but can also be widely applied to various large-span spatial structures such as stadiums, airport terminals, and convention centers. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart sensing and early warning system for large-span spatial structures based on BIM and an improved BP neural network, characterized in that, It includes a physical perception layer, a data integration and twin layer, an intelligent analysis and early warning layer, and an interactive application layer; The physical sensing layer consists of a variety of intelligent sensor nodes deployed in key parts of the large-span spatial structure, used to collect multi-dimensional physical quantities such as strain, acceleration, displacement, rotation angle and environmental parameters of the structure in real time. The data integration and twin layer includes a BIM-FEA integration engine, a monitoring data access gateway, and a dynamic digital twin model. The BIM-FEA integration engine realizes the automatic conversion of geometric, material, and boundary condition information from the BIM model to the finite element analysis model during the design phase, as well as the reverse mapping of FEA analysis results to the BIM model. The monitoring data access gateway receives, parses, and preprocesses multi-source heterogeneous monitoring data streams from the physical perception layer. The dynamic digital twin model realizes real-time, dynamic, semantic association and integrated storage of monitoring data, BIM geometric attributes, and FEA mechanical states through unique component identifiers. The intelligent analysis and early warning layer is the core computing unit, which has an improved physical-guided BP neural network prediction module, a damage identification module and an early warning rule engine built in. It receives time-series monitoring data from the data integration and twin layer and the corresponding FEA simulation benchmark data, uses the improved BP neural network to perform structural response prediction and abnormal pattern identification, and generates hierarchical early warning information by combining preset multi-level early warning thresholds and expert rules. The interactive application layer includes a visualization early warning plugin based on the BIM platform and a cloud-based collaborative management platform, which performs three-dimensional visualization rendering and interactive display of dynamic digital twin models, real-time monitoring data, and analysis and early warning results.

2. The intelligent sensing and early warning system for large-span spatial structures based on BIM and improved BP neural networks according to claim 1, characterized in that, The intelligent sensor nodes of the physical sensing layer include one or more of the following: fiber optic strain sensors, wireless accelerometers, laser displacement sensors, inclinometers, and temperature and humidity sensors.

3. The intelligent sensing and early warning system for large-span spatial structures based on BIM and improved BP neural networks according to claim 1, characterized in that, The BIM-FEA integration engine is based on industrial basic IFC standards or dedicated data interfaces to realize data conversion and mapping. The FEA analysis results include stress, strain, and displacement contour maps.

4. The intelligent sensing and early warning system for large-span spatial structures based on BIM and improved BP neural networks according to claim 1, characterized in that, The visualization and early warning plugin is developed in C# based on the Autodesk Revit API and includes a measurement point layout module, a measurement point search module, a real-time data module, and a historical data module. The measurement point layout module retrieves sensor IDs, creates sensor families, assigns them independent sensor IDs, and places them on the model surface to achieve sensor-component mapping. Clicking on a sensor on the model allows you to read real-time data. The measurement point search module iterates through sensor IDs to locate and highlight all measurement points. The real-time data module selects the detection factor and retrieves the real-time data from the measurement point sensors, synchronously updating the data transmission frequency. The historical data module allows you to select a date to view the historical data of several measurement points within a specified time period and displays the data trend.

5. The intelligent sensing and early warning system for large-span spatial structures based on BIM and improved BP neural networks according to claim 1, characterized in that, The cloud-based collaborative management platform adopts a microservice architecture to achieve lightweight loading of BIM models and 3D rendering on the web. Its core functions include statistical dashboards, 3D interaction, early warning and handling, and data analysis modules. The statistical dashboard displays the overall structural safety score, real-time early warning statistics, and an overview of the status of key sensors. The 3D interaction supports browser operations such as rotation, scaling, and sectioning of the lightweight BIM model, and clicking on components allows users to view full lifecycle data; the early warning and handling system can automatically push alarm information and initiate collaborative handling processes to form a closed-loop management system; the data analysis module supports trend analysis and correlation analysis of monitoring data.

6. A method for intelligent perception and early warning of large-span spatial structures based on BIM and an improved BP neural network, applied to the system described in any one of claims 1-5, characterized in that, Includes the following steps: S1. Constructing a dynamic digital twin model base for large-span structures: Based on design drawings and as-built data, establish a high-precision BIM model, generate a parametric finite element benchmark model through the BIM-FEA integration engine, add scalable attribute sets to key monitoring components in the BIM model, and form the initial digital twin model base. S2. Dynamic semantic fusion of multi-source monitoring data and BIM-FEA model: Based on structural mechanics analysis, the sensor network is optimized and deployed. The real-time sensor data stream is automatically matched and associated with the preset component ID in the BIM model through the monitoring data access gateway. The monitoring data is used to drive the finite element benchmark model for rapid simulation or model correction, and the simulation results are back-mapped to the corresponding component attributes in the BIM model. S3. Structural response prediction and anomaly detection based on temporal BP neural network: An improved three-layer BP neural network temporal prediction model is constructed, and a temporal window mechanism and a hybrid loss function are introduced to predict the sensor's measurement value at the next time step based on historical monitoring data, capturing the time dependence and nonlinear dynamic characteristics of the monitoring data. S4. Graded early warning generation and BIM visualization integration and release: Combining the damage probability output of the BP neural network prediction module, the over-limit situation of key indicator thresholds, and the correlation analysis results of multi-sensor data, graded early warning signals are generated. The early warning signals are bound to the BIM model related components through the visualization early warning plugin and displayed in three-dimensional visualization. At the same time, the early warning information is pushed to the cloud collaborative management platform. S5. Model update and system iterative optimization based on early warning feedback: Record the processing of early warning events and the results of structural inspection / maintenance to form a case library. Use feedback information to perform incremental learning optimization of the BP model, parameter correction of the finite element model, and adaptive adjustment of the early warning threshold.

7. The intelligent sensing and early warning method for large-span spatial structures based on BIM and improved BP neural network according to claim 6, characterized in that, The working mode of the BIM-FEA integration engine mentioned in step S1 includes: S1.1, forward conversion BIM->FEA: parsing the geometric entities, material properties and load case information of the BIM model, automatically converting them into finite element software input files based on the IFC general standard, and retaining the mapping relationship between component ID and BIM model; S1.2 Reverse Mapping FEA->BIM: Read the output result file of the finite element software, extract the stress, strain, displacement and mode shape data of the elements / nodes, and aggregate the result data onto the corresponding BIM components according to the mapping relationship. Store the data in the extended attribute set of the BIM model in the form of color gradient cloud map, numerical label or custom attribute.

8. The intelligent sensing and early warning method for large-span spatial structures based on BIM and improved BP neural network according to claim 6, characterized in that, The specific implementation of the dynamic semantic fusion of monitoring data and BIM model in step S2 includes: S2.1 Sensor Metadata Registration: Create a corresponding sensor family instance for each physical sensor in the BIM model and place it precisely at the monitoring point. The family instance contains sensor ID, type, range, installation direction, and the ID of the component to which it belongs, which are metadata attributes. S2.2 Real-time data flow association: Locate the corresponding sensor family instance and its attached parent component in the BIM model through the sensor ID, write the monitoring value into the dynamic attribute field of the parent component and add a timestamp. S2.3 Data-driven FEA rapid simulation: Using real-time monitored environmental parameters as input, it triggers parameterized finite element scripts to perform rapid simulation, obtains the theoretical structural response, and executes a reverse mapping process to update the theoretical state cloud map in the BIM model.

9. The intelligent sensing and early warning method for large-span spatial structures based on BIM and improved BP neural network according to claim 6, characterized in that, The prediction formula for the improved three-layer BP neural network time series prediction model described in step S3 is as follows: S3.1, The complete prediction formula is as follows: From input layer to output layer: Hidden layer activation: Hidden layer to output layer: Output layer activation: In the formula: Let i be the i-th input feature value; The connection weights between the i-th node in the input layer and the j-th node in the hidden layer are given. This is the bias term for the j-th node in the hidden layer; The weighted sum of the inputs to the j-th node in the hidden layer; Here is the activation function for the hidden layer; This is the activation output value of the j-th node in the hidden layer; The connection weight between the j-th node in the hidden layer and the node in the output layer; For output layer node bias terms; This is the weighted sum of the inputs for the output layer nodes; The activation function for the output layer; This represents the predicted output value of the neural network.

10. The intelligent sensing and early warning method for large-span spatial structures based on BIM and improved BP neural network according to claim 9, characterized in that, The training formula for the improved three-layer BP neural network time series prediction model described in step S3 includes: Output layer weight update: Hidden layer weight update: Output layer error: Hidden layer error In the formula: This is the true target value; This represents the predicted output value of the neural network. For output layer error sensitivity; Let be the error sensitivity of the j-th node in the hidden layer; This is the learning rate.

11. The intelligent sensing and early warning method for large-span spatial structures based on BIM and improved BP neural network according to claim 6, characterized in that, The graded warning signals mentioned in step S4 include multiple levels from "attention" to "serious", and the three-dimensional visualization warning methods include changing component color, flashing, pop-up window or one or more of these.

Images