Multimodal data-driven rail transit construction safety hazard identification method and system

Abstract

The application relates to a multi-modal data-driven rail transit construction safety hazard identification method and system, which solves the problem that multi-source data cannot realize deep cooperation, resulting in insufficient precision and real-time performance of hazard identification, and comprises the following steps: acquiring multi-modal data containing a BIM+GIS model, an unmanned aerial vehicle image stream and Internet of Things data; taking the BIM+GIS coordinates as a reference, and associating and binding data on an edge side and cleaning the data to a cloud side according to an edge side-cloud side architecture; correcting the cloud side to generate a standardized data set; identifying explicit and implicit hazards in stages, calculating a comprehensive risk index by using an entropy weight method, and generating a report to push a construction management terminal. The application has the following effects: breaking the multi-modal data barrier, realizing integrated and accurate identification of explicit and implicit hazards in rail transit construction, and fundamentally improving the accuracy and efficiency of construction safety control.

Description

Technical Field

[0001] This invention relates to the technical field of safety management in rail transit construction, and in particular to a multimodal data-driven method and system for identifying safety hazards in rail transit construction. Background Technology

[0002] Rail transit construction involves various work sites, including bridges, tunnels, and roadbeds. The construction environment is complex and constantly changing, and safety hazards are characterized by their hidden nature and strong correlation. Improper management can easily lead to safety accidents. Therefore, real-time and accurate identification of safety hazards during construction is a core requirement for ensuring project safety. With the application of information technology in the engineering field, traditional hazard identification methods relying on manual inspections are no longer sufficient to meet the safety management needs of rail transit construction, which involves a wide range of operations and complex conditions. There is an urgent need to leverage multi-source data fusion and intelligent algorithms to build an efficient safety hazard identification system, enabling dynamic control of safety risks throughout the entire construction process.

[0003] Currently, various information technologies have been introduced into the industry to explore and optimize hazard identification, forming a preliminary application foundation. Among them, drone technology is a commonly used auxiliary means in on-site monitoring, routinely used to assist manual inspections. For scattered bridges, roadbeds, and other construction sites along urban railway lines, drones capture images along preset flight paths, helping staff identify visible problems such as personnel not wearing protective equipment according to regulations and equipment temporarily placed outside safe areas. To some extent, this compensates for the limited coverage and low efficiency of manual inspections. However, these applications are mostly limited to image data collection and visual viewing, and the data is not deeply integrated with other management processes.

[0004] The core limitation of existing technologies lies in the failure to achieve deep collaboration among multi-source data, resulting in insufficient accuracy and real-time performance in hazard identification. Taking drone technology as an example, the image data it acquires is mostly stored and used independently, lacking effective correlation with basic information of the construction area (such as structural attributes of components) and real-time monitoring data (such as surrounding environmental parameters). It can only identify visible problems in isolation and cannot judge hidden risks caused by abnormal internal stress of components or excessive environmental indicators, or potential derivative risks behind visible problems (such as the possibility that improper placement of equipment may affect the safety of subsequent construction). It is difficult to cover the complex scenarios of safety hazards in rail transit construction. Summary of the Invention

[0005] In order to break down the barriers of multimodal data and achieve integrated and accurate identification of obvious and hidden hazards in rail transit construction, thereby fundamentally improving the accuracy and efficiency of construction safety management, this application provides a multimodal data-driven method and system for identifying safety hazards in rail transit construction.

[0006] Firstly, this application provides a multimodal data-driven method for identifying safety hazards in rail transit construction, employing the following technical solution:

[0007] A multimodal data-driven method for identifying safety hazards in rail transit construction includes:

[0008] Acquire multimodal data of the target rail transit project, including BIM+GIS models of geometric and mechanical properties of components at construction sites, continuous UAV image streams, personnel positioning data from IoT terminals, equipment operating parameters, and environmental monitoring data;

[0009] Based on the preset edge-cloud collaborative architecture, data is fused with the spatial coordinate system of the BIM+GIS model as the reference: After receiving the continuous image stream from the drone, personnel positioning data, equipment operating parameters, and environmental monitoring data, the edge device performs differentiated spatial association according to preset rules, completes the binding of the four types of data with the spatial elements of the model, and transmits them to the cloud after cleaning and compression.

[0010] The cloud calls up all attributes of the BIM+GIS model, corrects correlation deviations, integrates the data with the geometric and mechanical attributes and element numbers of the corresponding spatial elements, and generates a standardized dataset containing element numbers, spatial coordinates, time series, element attributes and monitoring data.

[0011] Perform hierarchical identification by calling a standardized dataset: use a preset algorithm to extract image stream features, combine personnel positioning data to identify visible hazards and mark them to corresponding spatial elements, and associate element numbers; for the marked elements and associated element numbers, extract relevant data and input them into a preset coupling model to deduce hidden hazards and output element risk values, which are mapped to risk levels;

[0012] The results of the two-level correlation verification are used to match spatial elements with their corresponding risk levels based on element numbers. The comprehensive risk index of each element number is calculated using the entropy weight method, and a report containing the hazard type, element number, comprehensive risk index, risk level and data source is generated and pushed to the construction management terminal.

[0013] By adopting the above technical solutions, this method can efficiently integrate multimodal data, solving the data fragmentation problem in traditional hazard identification; it accelerates data processing and fusion efficiency with an edge-cloud collaborative architecture, and ensures the accuracy of data correlation by relying on BIM+GIS benchmarks. Hierarchical identification can take into account both explicit and implicit hazards, and the comprehensive risk index calculated by the entropy weight method makes the assessment more objective. The final clear report is pushed to the terminal, which can help managers quickly grasp the hazard situation, improve the efficiency of safety management and control in rail transit construction, and reduce the risk of accidents.

[0014] Secondly, this application provides a multimodal data-driven system for identifying safety hazards in rail transit construction, employing the following technical solution:

[0015] A multimodal data-driven rail transit construction safety hazard identification system includes a memory, a processor, and a program stored in the memory and executable on the processor. When the program is loaded and executed by the processor, it implements the multimodal data-driven rail transit construction safety hazard identification method as described in the first aspect. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a multimodal data-driven method for identifying safety hazards in rail transit construction, according to an embodiment of this application. Detailed Implementation

[0017] The present application will be further described in detail below with reference to the accompanying drawings.

[0018] Reference Figure 1 This application discloses a multimodal data-driven method for identifying safety hazards in rail transit construction, comprising:

[0019] Step S1: Obtain multimodal data of the target rail transit project, including BIM+GIS model of geometric and mechanical properties of construction site components, continuous UAV image stream, personnel positioning data from IoT terminals, equipment operating parameters and environmental monitoring data.

[0020] The BIM+GIS model combines BIM (Building Information Modeling) and GIS (Geographic Information System) technologies. BIM is a 3D model that integrates information from the entire lifecycle of a building project, including the geometry, spatial relationships, geographic information, and attributes of building components. GIS is used to process and analyze geospatial data. The BIM+GIS model, by combining the two, can more comprehensively express the spatial and attribute information of the construction site. Continuous UAV image streams are sequences of image data acquired by continuously photographing the construction area using drones equipped with cameras and other devices. These images can reflect the real-time appearance of the construction site, including construction progress, material stacking, and personnel activities. IoT-based personnel location data utilizes IoT technology to obtain real-time location information of construction workers within the construction area by having them wear positioning devices (such as smart safety helmets with GPS and UWB positioning functions, positioning tags, etc.). Equipment operating parameters record the operating status parameters of various mechanical equipment (such as cranes, excavators, tunnel boring machines, etc.) during construction, including operating time, speed, load, oil temperature, and oil pressure. These parameters reflect the working status and normal operation of the equipment. Environmental monitoring data: Data obtained by monitoring environmental factors in the construction area, such as temperature, humidity, wind speed, dust concentration, noise intensity, and concentration of harmful gases. This data is crucial for assessing the safety of the construction environment.

[0021] Step S2: Based on the preset edge-cloud collaborative architecture, data is fused using the spatial coordinate system of the BIM+GIS model as the reference. After receiving the continuous image stream from the UAV, personnel positioning data, equipment operating parameters, and environmental monitoring data, the edge device performs differentiated spatial association according to preset rules, completes the binding of the four types of data with the spatial elements of the model, and transmits them to the cloud after cleaning and compression.

[0022] The edge-cloud collaborative architecture is a computing architecture that distributes data processing and analysis tasks between the edge (devices close to the data source) and the cloud (remote servers). The edge is responsible for initial data processing and filtering, while the cloud is responsible for complex data analysis and storage. A spatial coordinate system is a reference system used to define the location of points in space. In this application, the spatial coordinate system of the BIM+GIS model is a system combining a geographic coordinate system (such as WGS84) and the local coordinate system of the construction site, used to accurately describe the location of various elements within the construction area. Differentiated spatial association: Based on data type and application scenario, different spatial association rules and algorithms are used to bind different types of data with spatial elements in the BIM+GIS model.

[0023] The acquisition method and necessary process are described below:

[0024] 1. Edge-end data reception includes: 1. Continuous UAV image stream: The edge device receives continuous image stream data transmitted by the UAV via a wireless communication module. After reception, image processing algorithms (such as edge detection and feature extraction algorithms) are used to perform preliminary image processing, extract key feature points, and match them with spatial coordinates in the BIM+GIS model. For example, the SIFT (Scale Invariant Feature Transform) algorithm is used to extract feature points in the image, and then the ICP (Iterative Closest Point) algorithm is used to spatially match the feature points with geometric elements in the model. 2. Personnel positioning data: The edge device receives personnel positioning data transmitted from IoT terminals. Through positioning data parsing algorithms, the real-time location coordinates of personnel are converted into coordinates consistent with the spatial coordinate system of the BIM+GIS model. For example, coordinate transformation algorithms are used to convert GPS coordinates into local coordinate system coordinates of the construction site. 3. Equipment operating parameters: The edge device receives equipment operating parameters via wired or wireless communication modules. The equipment operating parameters are initially screened and normalized for subsequent association with equipment components in the BIM+GIS model. 4.

[0025] Environmental monitoring data: The edge device receives environmental monitoring data transmitted from environmental monitoring sensors. The data undergoes initial cleaning to remove outliers and noise. For example, a moving average algorithm is used to smooth the environmental monitoring data.

[0026] 2. The process of differential spatial association can be referred to steps S21 to S26, which will not be elaborated here.

[0027] 3. The data cleaning and compression process is as follows: 3.1 Data cleaning: Clean the correlated data to remove duplicate, abnormal, and noisy data. 3.2 Data compression: Compress the cleaned data to reduce the amount of data transmitted.

[0028] 4. The process of data transmission to the cloud is as follows: 4.1 Data encapsulation: The cleaned and compressed data is encapsulated into standardized data packets, containing information such as data type, timestamp, and spatial coordinates. 4.2 Data transmission: The data packets are transmitted to the cloud via secure communication protocols (such as TLS / SSL).

[0029] Step S3: The cloud calls up all attributes of the BIM+GIS model, corrects correlation deviations, integrates the data with the geometric and mechanical attributes and element numbers of the corresponding spatial elements, and generates a standardized dataset containing element numbers, spatial coordinates, time series, element attributes and monitoring data.

[0030] Among them, "full attributes" refers to all attribute information contained in the BIM+GIS model, including geometric attributes (such as the shape, size, and location of components) and mechanical attributes (such as material strength, elastic modulus, and load-bearing capacity). "Association deviation" refers to the deviation between the association result and the actual location or attribute caused by measurement errors, data processing errors, etc., during the data association process. "Standardized dataset" is a data set that has undergone cleaning, association, correction, and formatting, possessing a unified format and standard to facilitate subsequent analysis and processing.

[0031] The specific process for retrieving all attributes of the BIM+GIS model from the cloud and correcting correlation discrepancies can be found in steps S31 to S35, and will not be elaborated here. The remaining processes are as follows:

[0032] Integrate data with the geometric and mechanical properties of corresponding spatial elements:

[0033] Data fusion: This involves fusing the corrected multimodal data with the geometric and mechanical properties of the BIM+GIS model. For example, feature points in the UAV image stream are fused with the geometric properties (such as location and size) and mechanical properties (such as material strength) of model components to generate comprehensive data containing spatial coordinates, geometric properties, and mechanical properties.

[0034] Temporal data fusion: This involves integrating temporal data (such as personnel location data, equipment operating parameters, and environmental monitoring data) with spatial elements of the BIM+GIS model. For example, the timestamps of personnel location data are associated with the construction area in the model to generate spatial data containing time information. Example: Suppose a strain sensor is installed on a component in the construction area to monitor the stress on the component. The strain data (temporal data) collected by the sensor is fused with the mechanical properties (such as elastic modulus) of the component in the BIM+GIS model to generate comprehensive data containing timestamps, spatial coordinates, geometric attributes, and mechanical attributes.

[0035] The standardized dataset is generated as follows:

[0036] Data formatting: The merged data is formatted according to a unified format.

[0037] Data storage: Formatted data is stored as a standardized dataset. For example, data can be stored in a distributed storage system in the cloud (such as the Hadoop Distributed File System HDFS) for easier subsequent analysis and processing.

[0038] Step S4: Call the standardized dataset to perform hierarchical identification: Use a preset algorithm to extract image stream features, combine personnel positioning data to identify visible hazards and mark them to corresponding spatial elements, and associate element numbers; for the marked elements and associated element numbers, extract relevant data and input them into a preset coupling model, deduce hidden hazards and output element risk values, which are mapped to risk levels.

[0039] The process includes: Hierarchical identification: Safety hazards are categorized into different levels (e.g., visible and latent hazards) and identified and addressed separately. Visible hazards are obvious safety hazards that can be directly observed from data, such as personnel violations and equipment malfunctions. Latent hazards are potential safety hazards that are not easily observed directly, such as structural stability issues and material fatigue. Coupled models: Models used to analyze the interactions between multiple factors, comprehensively considering the impact of various factors on safety hazards.

[0040] The specific process of identifying visible hazards by combining personnel location data and marking them to corresponding spatial elements can be found in steps S41 to S46. The specific process of extracting relevant data, inputting it into a preset coupling model, deduce hidden hazards, output element risk values, and mapping them to risk levels can be found in steps S4a to S4e, and will not be elaborated here.

[0041] Step S5: Verify the results of the two levels, match spatial elements with corresponding risk levels based on element numbers, calculate the comprehensive risk index of each element number using the entropy weight method, generate a report containing hazard type, element number, comprehensive risk index, risk level and data source, and push it to the construction management terminal.

[0042] Among these, the following steps are included: **Association Verification:** Comparing and verifying the identification results of visible and hidden hazards to ensure their accuracy and reliability. **Element Number:** A unique identifier for each spatial element in the BIM+GIS model, used to distinguish different components or construction areas. **Entropy Weight Method:** A multi-attribute decision-making method based on information entropy, used to calculate the weight of each attribute to comprehensively assess the risk index.

[0043] Comprehensive Risk Index: A risk assessment indicator derived by comprehensively considering multiple factors, used to quantify the severity of safety hazards. Risk Level: A risk level classified according to the comprehensive risk index, typically divided into low risk, medium risk, and high risk.

[0044] The acquisition method and necessary process are described below:

[0045] 1. The two-level verification results are as follows: 1.1 Result Comparison: The identification results of visible and hidden hazards are compared and analyzed. For example, check whether the areas marked with visible hazards are spatially related to the results of the deduction of hidden hazards. If the risk value of the hidden hazard in the area of ​​the visible hazard is higher, the verification result is more reliable. 1.2 Consistency Check: The consistency of the identification results of visible and hidden hazards is verified through a consistency check algorithm. For example, if the visible hazard is a equipment failure, and the deduction result of the hidden hazard shows that the structural stability risk value of the area where the equipment is located is high, then the two have a certain correlation, and the verification result is reliable.

[0046] 2. Matching Spatial Elements with Corresponding Risk Levels Based on Element Numbers: 2.1 Element Number Matching: Match the identified visible and hidden hazards with the spatial element numbers in the BIM+GIS model. For example, mark the hazard on a specific component or construction area in the model and record the hazard type and risk level. 2.2 Risk Level Matching: Match the hazard with its corresponding risk level based on its risk value. For example, mark hazards with a risk value of 0-0.3 as low risk, 0.3-0.7 as medium risk, and 0.7-1 as high risk. Example: Assume a component has a hidden hazard risk value of 0.8, mark it as high risk, and label the hazard type and risk level on the corresponding component in the BIM+GIS model.

[0047] 3. Calculate the comprehensive risk index for each element number using the entropy weight method: 3.1 Data standardization: Standardize the risk values ​​of each hazard to a range between 0 and 1. For example, use the min-max normalization method to standardize the risk values. 3.2 Calculate information entropy: Calculate the information entropy of each hazard attribute using the information entropy formula. For example, calculate the information entropy for attributes such as hazard type, risk level, and data source, using the following formula: ; where p ij E is the standardized value of the j-th attribute of the i-th hidden danger, where n is the total number of hidden dangers. j This is the information entropy of the j-th attribute. Calculate the weights: Calculate the weights of each attribute based on its information entropy. The weight formula is as follows:

[0048] Where m is the total number of attributes.

[0049] Calculate the composite risk index: Using weights and standardized risk values, calculate the composite risk index for each element number. The formula is as follows: ;in, It is the comprehensive risk index of the i-th hidden danger. It is the standardized value of the j-th attribute of the i-th hidden danger.

[0050] 4. Generate a report: Generate a report that includes the type of hazard, element number, comprehensive risk index, risk level, and data source.

[0051] Perform differentiated spatial association according to preset rules to complete the binding of four types of data with model spatial elements, including:

[0052] Step S21: Use a preset feature extraction algorithm to extract dynamic target and static area feature points from the continuous image stream, combine BIM+GIS model parameters to perform bundle adjustment, establish a unified spatial benchmark with model spatial elements as anchor points, and achieve dynamic coordinate matching.

[0053] Among these, feature extraction algorithms are: computational methods used to identify and extract key information (feature points) from images. These feature points should possess uniqueness, distinguishability, and invariance under certain changes. Bundle adjustment is an optimization technique used to adjust image data acquired from multiple perspectives to minimize reprojection errors, thereby improving data consistency and accuracy. A unified spatial reference system ensures that all types of data (such as images, location data, etc.) are analyzed and processed in the same coordinate system. Dynamic coordinate matching updates and tracks the spatial position changes of objects in real time, ensuring that the current position of the object matches the spatial features in the model.

[0054] The acquisition method and necessary process are described below:

[0055] 1. Feature Extraction: Utilizing feature extraction algorithms (such as SIFT, SURF, or deep learning models) to analyze continuous UAV image streams, automatically identifying and extracting key feature points of dynamic targets (such as moving workers or vehicles) and static areas (such as building structures and construction equipment). These feature points, serving as carriers of key information in the images, will be used in the subsequent matching process with the BIM+GIS model.

[0056] 2. Integrating BIM+GIS Model Parameters: The extracted feature points are combined with the geometric and spatial parameters in the BIM+GIS model. The BIM model provides detailed building information, while the GIS model provides geospatial information. This combination ensures the consistency of the spatial location between the feature points and the corresponding spatial elements in the model.

[0057] 3. Bundle Adjustment: Bundle adjustment is performed on the extracted feature points to correct image distortion caused by factors such as shooting angle and lighting variations. This step helps improve the accuracy of matching feature points with spatial elements in the BIM+GIS model, ensuring data reliability.

[0058] 4. Establish a unified spatial benchmark: Using the spatial elements of the BIM+GIS model as anchor points, establish a unified spatial benchmark. This benchmark will serve as the basis for subsequent data association and analysis. A unified spatial benchmark ensures that all data sources (image streams, personnel positioning, equipment operating parameters, environmental monitoring data) are analyzed under the same coordinate system, thereby achieving data consistency and comparability.

[0059] 5. Achieve dynamic coordinate matching: On the established unified spatial benchmark, achieve dynamic coordinate matching between dynamic targets and static area feature points and model spatial elements. Dynamic coordinate matching allows the system to update and track the positional changes of objects within the construction area in real time, providing real-time spatial information for safety hazard identification.

[0060] Step S22: Based on the above benchmark, a preset filtering algorithm is introduced into the personnel positioning data to optimize the trajectory, anchor it to the corresponding element, associate the personnel type attribute, and compare it with the spatiotemporal target of the personnel in the image to form double verification.

[0061] The acquisition method and necessary process are described below:

[0062] 1. Trajectory Optimization: Pre-defined filtering algorithms (such as Kalman filtering and particle filtering) are used to process personnel location data acquired from IoT terminals to optimize personnel trajectories. These filtering algorithms reduce noise and errors in the location data, improving the smoothness and accuracy of the trajectory data.

[0063] 2. Anchoring to Corresponding Features: The optimized trajectory data is anchored to the corresponding spatial features in the BIM+GIS model. This step ensures that the personnel location data accurately corresponds to the geometric features in the model. For example, if a person is located near a specific construction area or equipment, their location data will be matched with the corresponding area or equipment in the model.

[0064] 3. Associate with job type attributes: Associate personnel location data with their job type attributes. This can be achieved through a pre-defined correspondence between job types and positioning devices. For example, personnel in different jobs may wear different types of positioning devices or tags, and this information can be used to identify the personnel's job type.

[0065] 4. Spatiotemporal Comparison: The optimized personnel trajectory data is compared spatiotemporally with the personnel targets captured in the UAV imagery. This step involves time synchronization and spatial matching. Through comparison, the accuracy of personnel positioning data can be verified, and possible abnormal behaviors or positional deviations can be identified.

[0066] 5. Create Dual Verification: By combining personnel location data and imagery data, dual verification of personnel activities and locations is created. This method improves the reliability and security of the data. For example, if location data shows that a person is working at a height, and imagery data also shows that the person is working at the corresponding location, the consistency and accuracy of the data can be confirmed.

[0067] Step S23: For the equipment operating parameters, parse the equipment ID and call the preset BIM equipment-work surface binding relationship to associate it with the current work element.

[0068] Among them, equipment operating parameters refer to various data generated during equipment operation, such as speed, load, temperature, and pressure. These parameters can reflect the operating status and performance of the equipment.

[0069] The acquisition method and necessary process are described below: 1. Parse Equipment ID: Extract the equipment ID from the equipment operating parameter data. This is key information for identifying and distinguishing different equipment. Equipment IDs are usually preset by the equipment manufacturer or coded and registered by the construction party before the equipment is put into use. 2. Retrieve BIM Equipment-Work Surface Binding Relationship: Based on the parsed equipment ID, find the corresponding equipment-work surface binding relationship in the BIM system. This binding relationship is preset during the construction project planning stage, defining the work tasks and positions of the equipment in different construction stages and on work surfaces. 3. Associate with Current Work Elements: Associate the equipment operating parameters with the current work elements. This means matching the actual operating data of the equipment with its preset work tasks and positions in the BIM model.

[0070] Step S24: Based on the above benchmark, the coverage of the environmental monitoring data is expanded using a preset interpolation algorithm, and an element-level environmental parameter field is generated by combining preset topological relationships.

[0071] Among them, the element-level environmental parameter field refers to the distribution of environmental parameters defined at each spatial element level in the BIM+GIS model, which is used to describe in detail the environmental conditions of each element in the construction environment.

[0072] The necessary processes are described below: 1. Data Collection and Preprocessing: Deploy an environmental monitoring sensor network to capture key environmental indicators such as temperature, humidity, wind speed, and dust concentration at the construction site in real time. Preprocess the collected data, including noise removal, missing value imputation, and outlier handling, to ensure data accuracy and usability. 2. Interpolation Algorithm Application: Select an interpolation algorithm suitable for the characteristics of the environmental monitoring data, such as Kriging interpolation or Inverse Distance Weighted (IDW) interpolation, to expand the coverage of the monitoring data. Apply the selected interpolation algorithm to estimate environmental parameter values ​​at unknown locations based on data from known monitoring points, generating a continuous environmental parameter field. 3. Topology Integration: Adjust the interpolated environmental parameter field by combining it with the topology of the construction site, such as equipment layout and construction area division. Utilize topology to ensure that the environmental parameter field matches the actual layout and operational logic of the construction site, improving the practicality and accuracy of the environmental parameter field. 4. Element-Level Environmental Parameter Field Generation: In the BIM+GIS model, generate a detailed environmental parameter field for each spatial element, including temperature, humidity, wind speed, and dust concentration fields.

[0073] Step S25: Using a preset multi-source data correlation evaluation model, calculate the spatial consistency and temporal correlation of related data for the same element, remove data with deviations exceeding the threshold, and verify the consistency of the correlation.

[0074] Among them, the multi-source data correlation assessment model is a model used to evaluate and quantify the degree of correlation between information from different data sources. Spatial fit refers to the consistency of spatial location information of the same element in different data sources. Temporal correlation refers to the synchronicity of time series data of the same element in different data sources. Data with deviation exceeding the threshold refers to data that exceeds the preset threshold in spatial fit or temporal correlation, which may need to be removed or further verified.

[0075] The acquisition method and necessary process are described below:

[0076] 1. Data collection: Data is collected from multiple sources, including BIM+GIS models, drone image streams, IoT terminals, and environmental monitoring equipment.

[0077] 2. Spatial fit calculation:

[0078] Method: Spatial analysis tools from a Geographic Information System (GIS) are used to calculate the spatial location deviation of the same feature from different data sources. Example: Suppose the center point coordinates of a construction device in the BIM model are (Xb, Yb), while the center point coordinates of the same device obtained from UAV image stream analysis are (Xi, Yi). Spatial fit can be calculated by determining the Euclidean distance D between the two points. To quantify.

[0079] 3. Time series correlation calculation:

[0080] Methods: Statistical analysis methods, such as Pearson correlation coefficient, were used to assess the correlation of time series data of the same element from different data sources.

[0081] Example: If both environmental monitoring data and equipment operation logs record changes in dust concentration within the construction area, the Pearson correlation coefficient between the two time series can be calculated. To assess the temporal correlation between them, where X and Y represent two time series, and It is their average value.

[0082] 4. Remove data with deviations exceeding the threshold:

[0083] Rules: Set thresholds for spatial fit and temporal correlation, such as a spatial distance threshold of 2 meters and a correlation coefficient threshold of 0.8. Method: Automatically mark and remove data points exceeding these thresholds. Example: If a data point has a spatial fit of 3 meters, exceeding the 2-meter threshold, or a temporal correlation of 0.6, below the 0.8 threshold, then this data point will be marked as an anomaly and removed from the analysis.

[0084] 5. Consistency check:

[0085] Methods: Cluster analysis or other statistical methods are used to assess the internal consistency of data point groups.

[0086] Example: K-means clustering analysis is used to divide personnel location data into different groups, each group representing a construction team. If the spatial consistency and temporal correlation of all data points within a group meet expectations, this indicates that the data associations are consistent.

[0087] Step S26: Based on the verification results, generate a spatiotemporal correlation map with model spatial elements as the core, and realize the three-dimensional binding of spatial coordinates, entity attributes and security semantics.

[0088] Based on the verification results, a spatiotemporal correlation map centered on model spatial elements is generated, realizing the three-dimensional binding of spatial coordinates, entity attributes, and security semantics, including:

[0089] Step S261: Initialize the map using the consistent model spatial features as the core nodes. The nodes integrate the spatial coordinates corresponding to the unified spatial benchmark and the geometric and mechanical attributes in the BIM+GIS model parameters. Based on the spatial coordinates, generate spatial adjacent edges between features, and based on the geometric and mechanical attributes, generate mechanically dependent edges to construct a relational network.

[0090] Among them, core nodes are: basic concepts in the graph, representing spatial elements of the model, such as building components, equipment, or construction areas. Spatial adjacency edges are edges connecting adjacent or close core nodes in physical space. Mechanically dependent edges represent edges indicating mechanical dependencies between core nodes due to structural or functional relationships. The relationship network is a network composed of core nodes and edges, describing the complex spatial and mechanical relationships between elements.

[0091] The acquisition method and necessary process are described below:

[0092] 1. Initialization of core nodes: Extract detailed information for each spatial feature from the BIM+GIS model, including spatial coordinates (X, Y, Z), geometric attributes (length, width, height), and mechanical attributes (material strength, elastic modulus). Each spatial feature is initialized as a core node in the map, carrying all its relevant attributes.

[0093] 2. Generate spatially adjacent edges:

[0094] Method: Spatial analysis algorithms (such as proximity analysis) are used to identify pairs of core nodes that are physically adjacent or close to each other. Algorithm Example: The Euclidean distance formula is used to calculate the distance between two nodes. If the distance is less than a preset threshold (e.g., 5 meters), a spatially adjacent edge is generated between them. Rule: A distance threshold, such as 5 meters, is set as the criterion for determining whether two nodes are spatially adjacent.

[0095] 3. Generate mechanically dependent edges:

[0096] Method: Analyze the geometric and mechanical properties of core nodes to determine their structural or functional dependencies. Algorithm Example: If the load of a beam member is directly supported by the column member below it, generate a mechanical dependency edge between the core nodes of these two members. This can be achieved by analyzing structural analysis data in the BIM model. Rule: Based on the structural analysis results, determine which nodes have mechanical dependencies.

[0097] 4. Construct a relational network: Integrate all core nodes and the edges between them (spatial adjacency edges and mechanically dependent edges) to construct a complete relational network. This network reflects not only spatial relationships but also mechanical dependencies, providing a foundation for subsequent data analysis and security assessment.

[0098] Step S262: Based on the coordinate association of spatially adjacent edges and the attribute association of mechanically dependent edges, the continuous image stream feature points, optimized personnel positioning trajectories, equipment operating parameters, and element-level environmental parameter fields that have passed the association degree evaluation are converted into dynamic data blocks in time sequence. These blocks are then associated with the corresponding core nodes through data attachment edges. The attribute records of these data attachment edges include the data acquisition timestamp and source reliability.

[0099] The dynamic data block is a collection of time-series data associated with specific model spatial features and changing over time. Data attachment edges are edges connecting the dynamic data blocks to core nodes in the spatiotemporal correlation graph, recording the data's source and timestamp. Source reliability is an indicator of data credibility or accuracy, used to assess data quality.

[0100] The acquisition method and necessary process are described below:

[0101] 1. Data Block Generation: Extract continuous image stream feature points, optimized personnel positioning trajectories, equipment operating parameters, and element-level environmental parameter fields from the data after correlation assessment. Organize these data into dynamic data blocks according to time series, with each data block containing relevant data over a period of time.

[0102] 2. Creation of Data Mounting Edges: For each dynamic data block, a data mounting edge is created based on its corresponding spatial features, linking the data block to the core node. The attributes of the data mounting edge include the data acquisition timestamp and source reliability, to record the data acquisition time and assess the data's credibility.

[0103] 3. Time-series data transformation: Collected environmental monitoring data, personnel location data, etc., are organized in chronological order to ensure data timeliness. Time series analysis methods, such as moving averages or time series models, are used to smooth the data to improve its accuracy.

[0104] Step S263: Call the preset safety rule library, combine the personnel type attributes and BIM equipment-work surface binding relationship, match the dynamic data with the semantic template, and generate real-time safety status labels.

[0105] Safety Rule Base: A database containing various predefined safety rules and standards used to assess the safety status of construction sites. Semantic Templates: A data structure used to match dynamic data with rules in the safety rule base to generate safety status labels. Real-time Safety Status Labels: Labels generated based on current data and the safety rule base, representing the immediate safety status of various elements at the construction site.

[0106] The acquisition method and necessary process are described below: 1. Access the safety rule base: Access the preset safety rule base, which contains safety standards and rules for different types of work, equipment, and working environments. 2. Combine with personnel type attributes: Match the personnel type attributes obtained from the personnel location data with the corresponding rules in the safety rule base. For example, if the rule base stipulates that electricians must wear insulated gloves during operation, the system will check whether the personnel identified as electricians in the location data have recorded information about wearing gloves. 3. Combine with BIM equipment-work surface binding relationship: Use the binding relationship between equipment and work surface in the BIM model to match the equipment operating parameters with the corresponding safety rules. For example, if the rule base stipulates that cranes must stop operating at a certain wind speed, the system will check the wind speed information in the environmental monitoring data and compare it with the crane's operating status.

[0107] 1. Matching Dynamic Data with Semantic Templates: Semantic templates are used to match collected dynamic data (such as personnel location, equipment operating parameters, environmental parameters, etc.) with rules in the security rule base. Semantic templates define the mapping relationship between data and rules, ensuring that data can correctly correspond to the corresponding security rules.

[0108] 2. Generate real-time safety status labels: Based on the matching results, generate real-time safety status labels for each spatial element. For example, if the dust concentration in a certain area exceeds the safety threshold, the system will generate a safety status label for that area indicating "high dust concentration risk".

[0109] Step S264: Based on the correlation parameters of spatially adjacent edges and mechanically dependent edges, calculate the environmental anomaly impact range for elements connected by spatially adjacent edges using a preset diffusion model, analyze the parameter chain effect for elements connected by mechanically dependent edges using a preset structural mechanics model, and label the derived risk semantics.

[0110] Among them, diffusion model: a mathematical model used to simulate and predict the propagation and impact range of environmental factors (such as dust, harmful gases, etc.) in space. Structural mechanics model: a mathematical model used to analyze and predict the response and behavior of building structures or components under stress. Chain reaction: changes in the state of other related elements caused by a change in the state of one element. Derived risk semantics: risk level or type labeled based on the analysis results of environmental impact range and chain reaction.

[0111] The methods and procedures are as follows:

[0112] 1. Calculation of the scope of environmental impact:

[0113] Methods: A pre-defined diffusion model, such as the Gaussian diffusion model, is used to predict the diffusion path and range of influence of environmental factors (such as dust and harmful gases).

[0114] Process: Collect environmental monitoring data, including pollutant concentration, wind speed, and wind direction. Utilize the Gaussian model, according to the formula... Calculate the pollutant concentration, where Q is the emission source intensity. This is a diffusion parameter. Areas where pollutant concentrations exceed safe thresholds are identified and marked as environmental risk areas.

[0115] 2. Structural Chain Effect Analysis:

[0116] Method: Use a pre-defined structural mechanics model, such as finite element analysis (FEA), to analyze the response of structural elements under stress.

[0117] Process: Extract the geometric and material properties of structural elements, as well as actual load data, from the BIM model. Apply the FEA model, using formulas... Calculate the stress distribution, where M is the bending moment and S is the section modulus.

[0118] Identify areas where the stress exceeds the material's yield strength and mark them as structural risk areas.

[0119] 3. Label the semantics of derived risks. For the specific process, please refer to steps S2641 to S2645.

[0120] 4. Update the correlation network: Update the calculated environmental impact range and the analysis results of the chain effect to the correlation network in real time.

[0121] Step S265: Establish a dynamic update mechanism: After receiving new verification data, locate the corresponding core node, synchronously update the dynamic data associated with its spatial coordinates and supplementary information of entity attributes, regenerate the real-time security status label and update the derived risk semantics, and maintain the real-time three-dimensional binding of spatial coordinates, entity attributes and security semantics.

[0122] The methods and procedures are as follows:

[0123] 1. Receive new verification data:

[0124] Method: The system is configured with a data interface to receive new verification data from environmental monitoring equipment, personnel positioning systems, equipment operation monitoring systems, etc. Process: The data interface can handle data input in different formats and transmit the data to the data processing module for further analysis.

[0125] 2. Locate the corresponding core node:

[0126] Method: Utilize identification information (such as device ID, personnel ID, etc.) in the data to locate the corresponding core nodes in the graph. Process: By querying the graph database, find the nodes related to the new data and prepare for data updates.

[0127] 3. Synchronously update dynamic data and entity attributes:

[0128] Method: Synchronize and update the newly received data with the existing data on the core nodes.

[0129] Process: For environmental monitoring data, update the node's environmental parameters, such as dust concentration and temperature. For personnel location data, update the node's personnel location and activity status. For equipment operation data, update the node's equipment status and operating parameters.

[0130] 4. Regenerate real-time security status labels:

[0131] Method: Rerun the evaluation model in the security rule base based on the updated data. Process: The evaluation model will analyze the updated data and generate new security status labels, such as "Safe," "Warning," or "Danger."

[0132] 5. Update the semantics of derived risks:

[0133] Method: Update the risk semantics in the graph based on the new safety status labels. Process: The updated risk semantics will be reflected in the visualization of the graph, providing managers with the latest risk information.

[0134] 6. Maintain the real-time performance of 3D rigging:

[0135] Method: Ensure synchronized updates of spatial coordinates, entity attributes, and safety semantics to maintain the real-time nature of the map. Process: Utilize real-time data streaming and event-driven update mechanisms to ensure that the information in the map remains consistent with the actual conditions at the construction site.

[0136] The semantics of derived risk annotation include:

[0137] Step S2641: Extract the association parameters of spatially adjacent edges and the association parameters of mechanically dependent edges.

[0138] Spatial adjacency edges refer to connections formed between two elements in the map due to their spatial proximity. Mechanically dependent edges refer to connections formed between two elements in the map due to their mechanical dependence. Association parameters include spatial coordinates, distance, direction, connection type, and load transfer coefficient, used to describe the association characteristics between elements.

[0139] 1. Extract the association parameters of adjacent edges in space:

[0140] Method: Using a GIS system, obtain the spatial coordinates of features and calculate the distance and direction between them. Process: For each pair of spatially adjacent features, record their coordinates, calculate the distance using the Euclidean distance formula, and determine their relative direction. Example: If feature A and feature B are adjacent in the map, the system will calculate the distance d between them. And record the direction information.

[0141] 2. Extract the correlation parameters of mechanically dependent edges:

[0142] Method: Extract the geometric and material properties of features from the BIM model, as well as connection types and load transfer factors. Procedure: For each pair of mechanically dependent features, extract their connection type (e.g., hinged, rigid) and load transfer factors, which describe how forces are transferred between structural features. Example: If features C and D are connected by a mechanically dependent edge, the system extracts their material properties and connection type, and records the load transfer factors for subsequent structural analysis.

[0143] Step S2642: For elements connected by spatially adjacent edges, input the abnormal data exceeding the safety threshold in the element-level environmental parameter field into the preset diffusion model, combine the spacing parameters of spatially adjacent edges, calculate the diffusion range, concentration decay coefficient and time-effect curve of environmental anomalies among related elements, and output the environmental anomaly impact boundary.

[0144] Among them, the concentration decay coefficient describes the rate at which pollutant concentration decreases with distance or time. The time-effect curve represents the change in pollutant concentration over time. The environmental anomaly impact boundary refers to the boundary of the area affected by pollutants or environmental factors, used to determine the affected region.

[0145] The acquisition methods and processes are described below: 1. Collect element-level environmental parameter data: Obtain real-time environmental parameter data, such as dust concentration and harmful gas concentration, from environmental monitoring equipment. Determine which data exceeds the preset safety threshold; this data will be used as input to the diffusion model. 2. Apply the diffusion model: Use a Gaussian diffusion model or other suitable diffusion model, taking the environmental parameter data exceeding the safety threshold as input. Combine this with the spacing parameters of adjacent spatial edges to calculate the diffusion path and impact range of pollutants in space. 3. Calculate the concentration decay coefficient and time-effect curve: Based on the output of the diffusion model, calculate the pollutant concentration decay coefficient with distance, which helps to understand the concentration changes of pollutants at different distances. Generate a time-effect curve to show the change of pollutant concentration over time, which is crucial for predicting and responding to environmental anomalies. 4. Output the environmental anomaly impact boundary: Based on the calculation results of the diffusion model, determine the specific boundaries of the environmental anomaly impact.

[0146] Step S2643: For elements connected by mechanically dependent edges, extract the geometric and mechanical properties of the core nodes and the equipment operation parameters in the dynamic data block, input the preset structural mechanics model, combine the load transfer coefficient of the mechanically dependent edge, analyze the chain effect caused by parameter anomalies, and output the mechanical response intensity.

[0147] Among them, the structural mechanics model is a mathematical model used to simulate and analyze the behavior and response of a structure under stress. The load transfer factor is a coefficient describing how forces are transferred between structural elements, affecting the stress analysis of the structure. The mechanical response intensity is a measure of the mechanical response of the structure under stress, such as stress and strain.

[0148] The specific process is as follows:

[0149] 1. Extract core node attributes and dynamic data:

[0150] Geometric and mechanical property acquisition: Retrieve precise dimensions and material property data of core nodes (such as beams, columns, slabs, etc.) directly from the BIM model database, including key parameters such as elastic modulus (E) and yield strength.

[0151] Dynamic data collection: Real-time equipment operating parameters are collected from IoT devices and sensors at the construction site. This data is stored in dynamic data blocks, including real-time load, vibration frequency, operating speed, etc.

[0152] 2. Application of structural mechanics models:

[0153] Model selection: Select a suitable finite element analysis (FEA) model that can handle complex structural analyses, such as nonlinear material behavior and large deformations.

[0154] Input data processing: Integrate the extracted core node attributes and equipment operating parameters from the dynamic data blocks to form the model input. This includes converting load data into a model-recognizable format and defining load transfer paths and coefficients based on mechanically dependent edges.

[0155] 3. Analyze the chain effect:

[0156] Simulation Analysis: Run the FEA model to simulate the structure's behavior under actual loading conditions. The model will calculate the stress and strain distribution for each node and element.

[0157] Chain reaction identification: Analyze the model output to identify chain reactions caused by abnormal parameters (such as overload), such as stress concentration areas or potential structural failure points.

[0158] 4. Output mechanical response intensity:

[0159] Results Extraction: Key mechanical response indicators, including maximum stress, strain distribution, and displacement, are extracted from the FEA model output.

[0160] Risk labeling update: Use these mechanical response data to update the risk labels in the map, and label each core node with real-time structural safety information, such as "high stress risk" or "structural stability warning".

[0161] Step S2644: Based on the environmental anomaly impact boundary output by the diffusion model and the mechanical response intensity output by the structural mechanics model, match the preset engineering risk semantic library to generate a preliminary version of the derived risk semantics for the related elements.

[0162] The necessary processes are as follows: 1. Collect model output: Obtain the environmental anomaly impact boundary from the diffusion model, including the pollutant diffusion range, concentration decay coefficient, and time-effect curve. Obtain the mechanical response intensity from the structural mechanics model, including key mechanical indicators such as maximum stress value and strain distribution. 2. Match the engineering risk semantic library: Match the output data of the diffusion model and the structural mechanics model with the entries in the engineering risk semantic library. The engineering risk semantic library is a predefined set of risk assessment standards, containing risk levels and descriptions under different environmental and structural conditions. 3. Generate the initial version of the derived risk semantics:

[0163] Based on the matching results, a preliminary risk semantic description is generated for each associated element. These descriptions include the risk type (e.g., "dust pollution", "structural overload"), risk level (e.g., "low", "medium", "high"), and potential impacts (e.g., "affects construction progress", "threatens personnel safety").

[0164] Step S2645: Combine the real-time security status label to revise the initial semantics, clarify the risk causes and affected objects, and form the final derived risk semantics.

[0165] The specific process is as follows: 1. Integrate real-time safety status tags: Collect real-time data from environmental monitoring, equipment monitoring, and other systems to generate safety status tags, such as "equipment overheating" and "area dust exceeding standards." These tags provide the current safety status of various elements at the construction site and are an important basis for revising risk semantics. 2. Revise the initial version of risk semantics: Compare and analyze the real-time safety status tags with the initial version of risk semantics generated in step S2644. Adjust the level and scope of impact in the risk semantics according to the new safety status tags to ensure the accuracy and timeliness of risk assessment. 3. Clarify risk causes and affected objects: Conduct a detailed analysis of each risk item to determine the specific cause of the risk, such as "ventilation equipment failure leading to dust exceeding standards." Identify the specific objects affected by the risk, such as "welding workers face health risks due to dust exceeding standards." 4. Quantify risk assessment: Quantify the risk, such as dividing the risk level into three levels: 1 (low), 2 (medium), and 3 (high), and setting specific quantitative standards for each level. For example, when the dust concentration exceeds the safety threshold by 20% but does not exceed 50%, the risk level is 2 (medium); when it exceeds 50%, the risk level is 3 (high).

[0166] 5. Formulate the final derived risk semantics: Based on the revised information and quantitative assessment results, formulate the final risk semantics, including the risk type, level, cause, and affected objects. For example, the final risk semantics might be: "The dust pollution risk level in Area A is 3 (high), caused by ventilation equipment failure, affecting the health of welding workers. It is recommended to repair the equipment immediately."

[0167] Accessing all attributes of the BIM+GIS model from the cloud and correcting correlation discrepancies includes:

[0168] Step S31: The cloud calls up all attributes of the BIM+GIS model and, combined with the unified spatial benchmark established at the edge, builds a benchmark framework for correlation deviation correction.

[0169] The specific process is as follows: 1. Access all attributes of the BIM+GIS model: Access the stored BIM+GIS model data in the cloud system to obtain all attribute information contained in the model. This attribute information includes, but is not limited to, geometric dimensions, material properties, spatial location, etc., providing basic data for subsequent deviation correction. 2. Combine with a unified spatial benchmark: Utilize the unified spatial benchmark already established at the edge to ensure the consistency of spatial coordinate systems across all data sources. A unified spatial benchmark is key to accurate data association and deviation correction, providing a common reference system for different data sources. 3. Build a benchmark framework: Construct a benchmark framework in the cloud, which will be used for deviation correction in subsequent steps. The benchmark framework needs to be able to process and integrate information from different data sources and provide a standardized processing flow. Assuming that at the construction site, the edge has already established a unified spatial benchmark through UAV image streams and personnel positioning data, the cloud system accesses the BIM+GIS model to obtain detailed geometric and attribute information of the construction site, such as the specific dimensions of buildings and the location of construction equipment. By combining a unified spatial benchmark at the edge, the cloud system builds a benchmark framework to correct the correlation deviations of image feature points, personnel trajectories, equipment operating parameters, and environmental parameter fields in subsequent steps.

[0170] Step S32: Based on the above benchmark framework, the correlation deviation between the feature points of the continuous image stream and the spatial elements of the model is processed: the RANSAC algorithm is used to remove mismatched feature points, and then perspective transformation is performed in combination with the geometric attributes of the BIM+GIS model to achieve accurate alignment between the spatial coordinates of the image feature points and the corresponding spatial elements of the model.

[0171] The necessary process is described below:

[0172] 1. Accessing the full attributes of BIM+GIS models from the cloud: Extracting complete attribute data of BIM+GIS models from the cloud database, including the precise coordinates, dimensions, and shapes of spatial elements.

[0173] 2. Establish a unified spatial benchmark: Ensure that all data sources (image streams, BIM+GIS models) are based on the same spatial coordinate system in order to make accurate comparisons and alignments.

[0174] 3. Use the RANSAC algorithm to remove mismatched feature points:

[0175] RANSAC algorithm: An iterative method for identifying the correct subset of data from a dataset containing a large number of outliers. This step identifies and removes mismatched feature points caused by factors such as changes in viewpoint and lighting conditions. The algorithm calculates model parameters by randomly selecting a small subset of data points and then evaluates the consistency of the remaining data points. This process is repeated until the best-fit model is found.

[0176] 4. Perform perspective transformation:

[0177] Perspective transformation: a geometric transformation used to map points in a two-dimensional image to three-dimensional space. In this step, the geometric attributes of the BIM+GIS model are used to perform perspective transformation on feature points in the image to achieve precise alignment with spatial elements in the model. The correspondence between feature points in the image and spatial elements in the model is calculated, and the perspective transformation formula is applied to adjust the position of the feature points so that they precisely match the three-dimensional coordinates in the model.

[0178] 5. Precise Alignment of Image Feature Points and Model Elements: Through the above processing, the system ensures that feature points in the image are precisely aligned with spatial elements in the BIM+GIS model in spatial coordinates, thereby correcting correlation deviations. Detailed Example: Suppose a drone captures an image of a construction site containing multiple identifiable feature points. First, the RANSAC algorithm is used to remove mismatched points caused by changes in viewpoint or lighting. Next, for the remaining feature points, perspective transformation is performed using the spatial element attributes in the BIM+GIS model. For example, if the 3D coordinates of a column in the model are known, the system will adjust the 2D representation of the column in the image to precisely align it with its 3D position in the model.

[0179] Step S33: To address the correlation deviation between personnel data and model spatial elements, based on the personnel positioning trajectory optimized at the edge, a dynamic time warping algorithm is introduced to align the temporal correlation relationship. At the same time, the work surface zoning attributes of the BIM+GIS model are combined to correct the spatiotemporal misalignment, thus forming a precise correlation between personnel trajectory and elements.

[0180] The necessary process is described as follows: 1. Retrieve full attributes of the cloud-based BIM+GIS model: Obtain detailed attributes of the BIM+GIS model from the cloud, including coordinates, dimensions, shape, and work area zoning information of spatial features. 2. Establish a unified spatial benchmark: Utilize the unified spatial benchmark established at the edge to ensure that personnel positioning data and model spatial features are compared and aligned in the same coordinate system. 3. Optimize personnel positioning trajectories: Use edge-based filtering algorithms (such as Kalman filtering) to optimize personnel positioning data, reduce noise and errors, and improve the accuracy of trajectory data. 4. Introduce Dynamic Time Warping (DTW): DTW algorithm: An algorithm for aligning time series data, capable of handling nonlinear relationships and velocity changes over time, ensuring consistency between trajectory data and model features in time. Alignment process: Compare the time series of personnel positioning trajectories with the work time periods defined in the BIM+GIS model. Use the DTW algorithm to find the optimal matching path between the two time series, minimizing the distance between them. Adjust the timestamps of the trajectory data to precisely align them with the work time periods defined in the model. 5. Correcting Spatiotemporal Misalignment: Combining the work area zoning attributes of the BIM+GIS model, correcting the spatiotemporal misalignment between personnel trajectories and features. Correction process:

[0181] Determine the spatial deviation between the worker's trajectory and the work area in the model. Based on the work area zoning attributes in the model, adjust the spatial position of the trajectory data to ensure a precise match with the work area in the model. For example, if a worker's trajectory shows them moving in an unauthorized area, the system will correct it according to the model attributes to ensure the trajectory data reflects the worker's activity within the correct work area. 6. Establish precise associations:

[0182] Through the above steps, a precise correlation is formed between personnel trajectories and model spatial elements, ensuring the accuracy and reliability of the correlation.

[0183] Step S34: For the correlation deviation between equipment operating parameters and model space elements, based on the preset BIM equipment-work surface binding relationship, the mapping relationship between equipment ID and work elements is parsed, and the attribute correlation deviation is corrected through parameter matching algorithm to ensure the consistency of the correlation between equipment operating parameters and corresponding elements.

[0184] The necessary processes are described below: 1. Retrieve all attributes of the BIM+GIS model: Obtain complete attribute information provided by the BIM+GIS model from the cloud. This information includes the spatial location of the equipment, design parameters, expected operating status, etc. 2. Establish a unified spatial benchmark: Ensure that the equipment operation data and the spatial data of the BIM+GIS model are in the same spatial coordinate system for accurate data correlation and comparison. 3.

[0185] 4. Resolve the mapping relationship between equipment IDs and work elements: Use the equipment ID to find the corresponding work surface or spatial element in the BIM model and resolve the preset binding relationship between them. 5. Apply parameter matching algorithms: Use parameter matching algorithms, such as fuzzy matching or nearest neighbor matching, to identify the correlation deviation between equipment operating parameters and model spatial elements. The algorithm compares the actual operating parameters of the equipment with the expected operating parameters of the model, identifies the deviation, and corrects it. 6. Correct attribute correlation deviations:

[0186] Based on the results of the matching algorithm, the correlation between the equipment operating parameters and the model space elements is adjusted to ensure consistency between the two.

[0187] Step S35: To address the correlation deviation between the feature-level environmental parameter field and the spatial features of the model, based on the spatial topology relationship of the BIM+GIS model in the benchmark framework, a spatial interpolation algorithm is used to optimize the spatial distribution of the parameter field. Combined with the preset topology relationship, the correlation deviation between environmental data and the spatial range of features is corrected, so that the environmental parameter field accurately covers the corresponding features.

[0188] The necessary process is described below:

[0189] 1. Utilize the spatial topology relationships of the BIM+GIS model: Extract spatial topology relationships from the BIM+GIS model. These relationships describe spatial relationships such as adjacency, inclusion, or connection between spatial elements.

[0190] 2. Employ spatial interpolation algorithms: Select appropriate spatial interpolation algorithms, such as Kriging interpolation, inverse distance weighted interpolation, or multiple linear regression interpolation, to optimize the spatial distribution of the environmental parameter field. The interpolation algorithm estimates environmental parameter values ​​at unknown locations based on known environmental monitoring data points, generating a continuous environmental parameter field.

[0191] 3. Optimize the spatial distribution of the parameter field: Apply spatial interpolation algorithms to optimize the spatial distribution of the environmental parameter field based on the spatial topology of environmental monitoring data and the BIM+GIS model. Ensure that the environmental parameter field reflects the actual environmental conditions of the construction site, such as temperature, humidity, wind speed, and dust concentration.

[0192] 4. Correcting the correlation deviation between environmental data and feature spatial extent: Based on the preset spatial topology, correcting the correlation deviation between the environmental parameter field and the spatial features of the model. Adjusting the environmental parameter field to accurately match the spatial feature extent defined in the BIM+GIS model.

[0193] By combining preset topological relationships to correct the correlation deviation between environmental data and the spatial extent of elements, the environmental parameter field accurately covers the corresponding elements, including:

[0194] Step S351: Extract the spatial topological relationships of the BIM+GIS model in the benchmark framework, including the inclusion, adjacency, and subordination relationships between elements, and construct a topological rule library for element-environment association by combining preset engineering topological rules.

[0195] The necessary process is as follows: 1. Extract spatial topological relationships: Extract spatial topological relationships between features from the BIM+GIS model, including containment, adjacency, and dependency relationships. These relationships define the spatial logical connections between features. 2. Construct a topology rule base: Combine preset engineering topology rules to construct a topology rule base to describe the association rules between features and the environmental parameter field. The rule base should include how to handle spatial relationships between features and how to optimize the coverage of the environmental parameter field based on these relationships. 3. Define association rules: Define association rules between features and the environmental parameter field. For example, if one feature is contained by another feature, the environmental parameter field should cover the containment relationship accordingly. The rules should also include how to handle adjacency relationships between features and how to adjust the boundaries of the environmental parameter field based on adjacency relationships.

[0196] Step S352: Analyze the 3D coverage of the optimized feature-level environmental parameter field, and extract the geometric boundaries of the corresponding spatial features from the BIM+GIS model to generate a comparison matrix between the parameter field boundary and the feature boundary.

[0197] The necessary processes are as follows: 1. Analyze the 3D coverage: For each feature-level environmental parameter field, analyze its coverage in 3D space. This includes determining the spatial distribution of the parameter field, such as the specific distribution area of ​​environmental factors like dust concentration and temperature at the construction site. 2. Extract geometric boundaries: Extract the precise geometric boundaries of each spatial feature from the BIM+GIS model. These boundaries define the location and shape of the feature in 3D space. 3. Generate a comparison matrix: Create a comparison matrix to compare the coverage of the environmental parameter field with the geometric boundaries of the spatial features. Each element in the matrix represents the spatial relationship between the parameter field and the boundary of a specific feature. 4. Identify deviations: Identify the deviations between the environmental parameter field and the boundary of the spatial features using the comparison matrix. This may include areas where the parameter field extends beyond the feature boundary (redundant areas) or areas where the parameter field does not cover the feature (missing areas).

[0198] Step S353: Identify and process the deviation type based on the comparison matrix: For redundant areas in the parameter field that exceed the feature boundary, trim them according to the feature subordination relationship in the topology rule base; For feature areas not covered by the parameter field, extract data from neighboring parameter fields based on the association relationship between adjacent features, and supplement the coverage through weighted interpolation.

[0199] The necessary processes are as follows: 1. Identify the type of deviation: Using the comparison matrix generated in step S352, identify the type of deviation between the environmental parameter field and the spatial feature boundary, including redundant areas where the parameter field exceeds the feature boundary and feature areas not covered by the parameter field. 2. Process redundant areas: For redundant areas where the parameter field exceeds the feature boundary, trim them according to the feature dependency relationships in the topology rule base. Trimming method: Use geometric operations, such as intersection or difference, to adjust the boundary of the parameter field to match the feature boundary. 3. Supplement uncovered areas: For feature areas not covered by the parameter field, extract data from neighboring parameter fields based on the association relationships of adjacent features. Interpolation method: Use a weighted interpolation algorithm to supplement the environmental parameters of the uncovered areas based on the data and spatial relationships of neighboring parameter fields.

[0200] Step S354: Introduce the spatial fit calculation model to quantify the matching degree between the corrected parameter field and the spatial range of the elements. If the matching degree is lower than the preset threshold, return to the previous step to re-optimize the topological association rules and interpolation parameters; if the matching degree is higher than or equal to the threshold, proceed to the next step.

[0201] The necessary process is as follows: 1. Introduce a spatial fit calculation model: Use a specific calculation model to quantify the degree of fit between the environmental parameter field and the spatial feature boundary. This model may be based on geometric similarity, coverage, or other relevant metrics. 2. Calculate the matching degree: Using the above model, calculate the matching degree between the corrected environmental parameter field and each relevant spatial feature. This involves comparing the actual coverage of the parameter field with the expected boundary of the feature. 3. Set a preset threshold: Determine a preset matching degree threshold as a standard for judging whether the matching between the parameter field and the feature spatial range is acceptable. 4. Evaluation and adjustment: If the calculated matching degree is lower than the preset threshold, it indicates that the matching between the parameter field and the feature spatial range is not accurate enough, and it is necessary to return to step S353 to re-optimize the topological association rules and interpolation parameters. If the matching degree is higher than or equal to the threshold, it indicates that the matching between the parameter field and the feature spatial range is acceptable, and the next step can be performed.

[0202] Step S355: Associate the timestamps of dynamic data blocks, perform consistency verification on the parameter field coverage of the same element at different time series, correct the time series correlation deviation caused by dynamic changes in construction, and ensure the coverage accuracy of the spatiotemporal dimension.

[0203] The necessary processes are described below: 1. Associate timestamps with dynamic data blocks: Add a timestamp to each environmental monitoring data point to ensure the accurate recording of the data's temporal attributes. Associate these timestamps with dynamic data blocks, which contain environmental parameter data that changes over time. 2. Perform temporal correlation analysis: Conduct a detailed comparison of the environmental parameter fields of the same spatial feature at different time points. Analyze how the parameter fields change over time and identify any trends or patterns that do not conform to expectations. 3. Consistency verification: Verify whether the environmental parameter fields at different time points are consistent with the feature boundaries in the BIM+GIS model in terms of spatial coverage. Check for any deviations in the time series, which may be due to changes in construction activities or abrupt changes in environmental conditions. 4. Correct temporal correlation deviations: For identified deviations, analyze their causes, which may include adjustments to the construction schedule, changes in equipment location, or changes in environmental conditions. Based on the analysis results, adjust the coverage of the environmental parameter fields to match the latest construction status and environmental conditions. 5. Ensure spatiotemporal coverage accuracy: Through iterative verification and correction, ensure the coverage accuracy of the environmental parameter fields in both time and space dimensions. This process may require multiple iterations until the coverage of the environmental parameter field is consistent with the spatial feature boundary at all points in time.

[0204] Step S356: Output the feature-level environmental parameter field that precisely matches the coverage area and the feature spatial boundary, and record the basis for correction.

[0205] The necessary process is as follows: 1. Output a precisely matched environmental parameter field: After completing all necessary corrections, the final output environmental parameter field should be completely consistent with the spatial boundaries of the features in the BIM+GIS model. These parameter fields reflect the actual environmental conditions of the construction site, such as dust concentration, temperature, humidity, etc., and the coverage area precisely corresponds to each construction feature. 2. Record the basis for corrections: Record the specific reasons for each correction and the methods used, including the algorithms, interpolation strategies, and clipping rules used. Save all intermediate and final data during the correction process for future review and verification.

[0206] Image stream features are extracted using a pre-defined algorithm, and visible hazards are identified and annotated to corresponding spatial elements based on personnel location data, including:

[0207] Step S41: Extract the basic data associated with the labeled features, including continuous UAV image streams, real-time personnel positioning data, and BIM+GIS coordinate information of the corresponding model spatial features.

[0208] The necessary processes are described below: 1. Collect UAV image stream data: Collect continuous image stream data from the construction site using UAVs. This data provides real-time visual information about the construction site. 2. Obtain real-time personnel location data: Collect real-time location data through positioning devices worn by construction workers (such as smart safety helmets, positioning badges, etc.). This data reflects the dynamic location information of the personnel. 3.

[0209] 4. Extract coordinate information from the BIM+GIS model: Extract precise coordinate information for each spatial element from the BIM+GIS model, including X, Y, and Z coordinates. This coordinate information is crucial for associating imagery data and personnel positioning data with the model's spatial elements. 5. Integrate and link data: Integrate UAV imagery stream data, personnel positioning data, and coordinate information from the BIM+GIS model to form a complete dataset for subsequent hazard identification and labeling. 6. Data preprocessing: Perform necessary preprocessing on the collected data, such as image format conversion and filtering of positioning data, to improve data quality and the accuracy of subsequent processing.

[0210] Step S42: The continuous image stream of the UAV is processed by a preset image feature extraction algorithm. First, redundant frames are removed by the inter-frame difference method, and then the target is identified by the preset target detection model. The pixel coordinates and category labels of each target are output.

[0211] The necessary processes are as follows: 1. Inter-frame differencing: Using image processing techniques, inter-frame differencing is performed on the continuous image stream to identify and highlight parts of the image that change over time. This method can effectively remove static backgrounds, retaining only dynamic changes, such as moving people or equipment. 2. Application of object detection models: Pre-set object detection models, such as YOLO, Faster R-CNN, and SSD, are used to analyze the processed image to identify and locate objects in the image. Object detection models can identify various types of objects and output pixel coordinates and category labels for each identified object. 3.

[0212] Output target information: After the model is processed, the pixel coordinates and category label of each identified target are output.

[0213] Step S43: The real-time positioning data of personnel is preprocessed using a preset positioning optimization algorithm to optimize the original trajectory to eliminate signal drift and output the real-time spatial coordinates of personnel that are consistent with the BIM+GIS model coordinate system.

[0214] The execution process is as follows: 1. Collect personnel positioning data: Collect real-time personnel positioning data from on-site positioning devices (such as GPS, RFID, Bluetooth beacons, etc.). 2. Apply positioning optimization algorithms: Preprocess the collected raw positioning data using filtering algorithms (such as Kalman filtering) to eliminate noise and correct errors. Use smoothing or clustering algorithms to optimize the trajectory to reduce unrealistic movements caused by signal drift. 3. Eliminate signal drift: Identify and correct positioning deviations caused by signal drift, such as multipath effects or non-line-of-sight errors. Improve the stability and accuracy of positioning data through algorithm enhancement. 4. Output consistent coordinates: After optimization, the output personnel location data will be consistent with the coordinate system used in the BIM+GIS model. This step ensures that the positioning data can be directly applied to the BIM+GIS model for subsequent analysis and decision-making. Example: Suppose a worker on a construction site wears a positioning device that records his movement trajectory. The raw data may contain errors due to signal reflection or obstruction. Apply algorithms such as Kalman filtering to process this data, eliminate errors, and optimize the trajectory. The processed data output is precise spatial coordinates of the personnel, which can be directly mapped to the corresponding locations in the BIM+GIS model.

[0215] Step S44: Establish a connection between the target information extracted from the image and the spatial elements of the model using a preset coordinate mapping method, convert the target pixel coordinates of the image into absolute coordinates of the model, and match the preset spatial element type with the coordinate range.

[0216] The imagery includes the following components: Image target: Objects identified in the UAV imagery, such as workers and equipment. Pixel coordinates: The position of the target in the imagery, usually with the top left corner of the image as the origin. Absolute coordinates: The actual coordinates of the target's location on the construction site, typically consistent with the coordinate system used in the BIM+GIS model. Coordinate mapping: The process of converting the pixel coordinates in the imagery into the absolute coordinates of the actual construction site.

[0217] The specific process is as follows: 1. Extracting target information from images: Image processing algorithms (such as Canny edge detection, Sobel operator, etc.) are used to extract target features from images captured by the drone. Target detection algorithms (such as YOLO, Faster R-CNN) are applied to identify targets and label their pixel coordinates and category labels. 2. Coordinate mapping method: The intrinsic parameters (focal length, principal point, etc.) and extrinsic parameters (geographic location, attitude, etc.) of the drone camera are determined. These parameters are used to convert pixel coordinates into absolute coordinates. The following formula is used for coordinate transformation: ; ;in, It is the X-pixel position of the target in the image. It is the focal length of the camera on the X-axis. These are the world coordinates of the camera's principal point on the X-axis; It is the Y-pixel position of the target in the image. It is the camera's focal length on the Y-axis. 1. The world coordinates of the camera's principal point on the Y-axis. 2. Convert to absolute coordinates: Convert the target pixel coordinates in the image to absolute coordinates of the construction site, ensuring accuracy and consistency in the coordinate transformation. 3. Match spatial feature types: Based on the converted absolute coordinates and the preset coordinate range, match the coordinates with preset spatial feature types in the BIM+GIS model. For example, if the converted coordinates fall within a specific construction area, associate them with the spatial features of that area.

[0218] Step S45: Combine the spatial association results with the preset explicit hazard rule base to make a judgment. If the personnel coordinates trigger the danger zone range, the equipment coordinates exceed the preset position deviation threshold, or the component features match the damage label, it is judged as an explicit hazard.

[0219] The general process is as follows: 1. Combine spatial association results: Utilize the association between image targets and model spatial elements established in step S44 to obtain the absolute coordinates and category label of each target. 2. Apply the explicit hazard rule base: The explicit hazard rule base contains a series of predefined rules used to identify and determine potential hazards at the construction site. These rules may include personnel location restrictions, equipment operation specifications, and structural integrity standards. 3. Perform judgment: For each associated target, compare its coordinates and category with the rules in the explicit hazard rule base. Check for the following: Whether personnel coordinates have entered a predefined danger zone; Whether equipment coordinates have exceeded a preset position deviation threshold; Whether component features match the damage label. 4. Mark explicit hazards: If a target violates any rule in the explicit hazard rule base, mark it as an explicit hazard. Record detailed information about the hazard, including type, location, timestamp, and related target coordinates.

[0220] Step S46: Mark the identified visible hazards to the corresponding model space elements and record the hazard type, timestamp, and target coordinates.

[0221] The specific process is as follows: 1. Data Integration and Processing: Obtain explicit hazard information from step S45, including personnel coordinates, equipment coordinates, and component characteristics. Use spatial data processing technologies, such as coordinate transformation and format unification, to ensure the compatibility of hazard information with the BIM+GIS model. 2. Accurate Hazard Labeling: Utilize the APIs or data interfaces provided by the BIM+GIS model to label hazard information onto the corresponding spatial features of the model. Use Geographic Information System (GIS) functions, such as spatial join or spatial query, to automatically match hazard coordinates with model features. 3. Recording Detailed Information: Create hazard records in the database, including the following fields: Hazard type (e.g., "Personnel Unauthorized Entry", "Equipment Exceeding Limits", "Structural Damage"). Timestamp (the specific date and time of the hazard occurrence). Target Coordinates (the precise geographical location of the hazard occurrence). Use Structured Query Language (SQL) or NoSQL databases to store and manage these records. 4. Updating the Model and Database: Update the labeled hazard information to the BIM+GIS model and related databases in real time. Ensure that the model and database reflect the latest safety status of the construction site and support dynamic safety management.

[0222] For the labeled elements and related element numbers, relevant data is extracted and input into a preset coupling model to deduce hidden risks and output element risk values, which are mapped to risk levels, including:

[0223] Step S4a: Extract the associated data of the labeled elements, including explicit hazard labeling information, geometric and mechanical properties of the BIM+GIS model, equipment operating parameters, and element-level environmental monitoring data.

[0224] Step S4b involves processing the extracted associated data using a pre-defined standardization algorithm: unifying the dimensions and time granularity of equipment operating parameters, filling in the time-series missing values ​​in environmental monitoring data, converting explicit hazard labeling information into quantitative triggering factors, and outputting a standardized multi-dimensional parameter set.

[0225] The necessary processes are as follows: 1. Unify the dimensions and time granularity of equipment operating parameters: Convert different equipment operating parameters (such as speed, load, vibration, etc.) into a unified dimension and time granularity (such as seconds, minutes) to eliminate the influence of different equipment and time scales. 2. Fill in the missing time series values ​​in environmental monitoring data: Use interpolation algorithms (such as linear interpolation, nearest neighbor interpolation, or more complex time series prediction models) to fill in the missing values ​​in environmental monitoring data to ensure data integrity. 3. Convert explicit hazard labeling information into quantitative triggering factors: Convert qualitative explicit hazard labeling information (such as "personnel violation" or "equipment failure") into quantitative triggering factors, which can be numerical representations of the probability or severity of hazard occurrence. 4. Output a standardized multi-dimensional parameter set: Integrate the processed data into a standardized multi-dimensional parameter set, where each dimension represents a specific attribute or indicator (such as equipment operating status, environmental conditions, etc.).

[0226] Step S4c involves mapping the standardized multi-dimensional parameter set to the geometric and mechanical properties of the BIM+GIS model: binding equipment operating parameters to the mechanical parameters of the corresponding work surface components, associating environmental monitoring data with component material properties, and generating an element-parameter association matrix.

[0227] The necessary processes are as follows: 1. Mapping equipment operating parameters to component mechanical parameters: Mapping equipment operating parameters (such as load, speed, vibration, etc.) to the corresponding components on the working surface in the BIM model. Based on the equipment type and the specific conditions of the working surface, associating the operating parameters with the mechanical properties of the components (such as strength, stiffness, etc.). 2. Associating environmental monitoring data with material properties: Associating environmental monitoring data (such as temperature, humidity, wind speed, etc.) with the material properties of the components. Based on the influence of environmental factors on material properties, such as the influence of temperature on the elastic modulus of steel, establishing the association between environmental data and material properties. 3. Generating a feature-parameter association matrix: Creating a matrix where rows represent different spatial features (such as different components) and columns represent different parameters (such as equipment operating parameters and environmental monitoring data). Each entry in the matrix represents the strength or degree of association between a specific feature and a specific parameter.

[0228] Step S4d: Input the element-parameter association matrix into the preset coupling model and use the preset fusion algorithm for hierarchical quantification: Calculate the cumulative deformation and stress over-limit frequency by combining the component mechanical parameters and equipment operating parameters using the preset mechanical calculation algorithm; infer the corrosion depth and durability decay rate by associating material properties with environmental monitoring data using the preset environmental erosion algorithm; and then sum according to the preset weighted fusion rules to output the comprehensive quantitative value of hidden dangers as the element risk value.

[0229] The general process is as follows: 1. Input element-parameter correlation matrix: Input the element-parameter correlation matrix generated in step S4c into the coupled model. This matrix contains the mechanical parameters of the components, equipment operating parameters, and environmental monitoring data. 2. Mechanical calculation algorithm (e.g., Finite Element Analysis (FEA)): Method: Apply finite element analysis (FEA) to simulate the response of the components under actual loads. Process: Use FEA software, such as ANSYS or ABAQUS, to build a three-dimensional model of the components. Input the geometric dimensions, material properties, and boundary conditions of the components. Apply equipment operating parameters, such as periodic loads, to the model to simulate stress and deformation under load. Calculate the cumulative deformation and stress over-limit frequency to assess the stability and safety of the structure. Example: Assume a support beam is subjected to periodic loads. Use the FEA model to calculate the deformation and peak stress of the beam over a certain period of time. If the calculation results show that the maximum stress value exceeds the yield strength of the material, it indicates that the beam is at risk of structural failure. 3. Environmental erosion algorithm (e.g., uniform corrosion model): Method: Use environmental impact assessment models, such as uniform corrosion models, to predict the durability of materials. Process: Based on environmental monitoring data (such as humidity and chemical concentration) and material properties (such as corrosion resistance), the corrosion rate formula is applied to calculate the corrosion depth. The durability degradation rate of the material is assessed, and the service life of the material is predicted. Example: For steel structures exposed to high humidity environments for extended periods, a uniform corrosion model is used to assess their corrosion rate and remaining lifespan. If the calculation results show that the corrosion depth exceeds a preset safety threshold, it indicates that the steel structure requires early maintenance. 4. Layered Quantification of Fusion Algorithm: Method: Weighted summation or other multi-criteria decision-making methods are used to integrate mechanical and environmental data. Process: According to preset weighted fusion rules, the mechanical calculation results and environmental erosion results are combined to output a comprehensive quantitative value. Weighted fusion considers the contribution of different parameters to the overall risk, ensuring the comprehensiveness and accuracy of the risk assessment. Example: If a concrete member is subjected to both cyclic loads and long-term exposure to a corrosive environment, the coupled model will combine these two factors to give a comprehensive risk score. This score can be a value between 0 and 1, where 1 represents the highest risk. This score is used to determine whether emergency measures or long-term maintenance plans are needed. 5. Outputting Comprehensive Quantitative Values ​​of Hidden Hazards: Method: Outputting the risk score or hazard index for each spatial element. Process: Mapping the quantitative values ​​to preset risk levels, such as low, medium, and high risk. Example: For the above concrete component, if the comprehensive score is high, it indicates that it may have hidden risks of structural failure or reduced durability, requiring immediate action.

[0230] Step S4e: Match the comprehensive quantitative value to the corresponding level using a preset risk level mapping rule to determine the risk level of the element.

[0231] The necessary process is as follows: 1. Adopt preset risk level mapping rules: Use predefined risk level mapping rules to convert the comprehensive quantitative value into a risk level. These rules are based on historical data, industry standards, or expert knowledge. 2. Match corresponding levels: Compare each comprehensive quantitative value with the risk level mapping rules to determine its corresponding risk level. Risk levels are usually divided into several levels, such as "low," "medium," and "high," each level corresponding to different risk management and response measures. 3. Determine element risk value: Based on the mapping results, assign a risk value to each spatial element. This value reflects the element's potential risk level. The risk value can be used for subsequent risk assessment, decision support, and resource allocation. Example details: Suppose the coupled model calculates a comprehensive quantitative value of 0.75 for a support column at a construction site. According to the preset risk level mapping rules, 0.75 may map to a "medium" risk level. The system marks the risk value of this support column as "medium" and records it in the database, while also marking it in the BIM+GIS model with an appropriate color or icon. Based on this risk value, managers can decide whether further inspection, reinforcement, or changes to the construction plan are needed for the support column. 5. Output Risk Report: Generate a report containing the risk levels of all spatial elements for construction management personnel to analyze and make decisions. The report should include risk values, timestamps, spatial coordinates, and relevant suggested countermeasures.

[0232] The multimodal data-driven method for identifying safety hazards in rail transit construction also includes a step of deducing hidden hazards, as detailed below:

[0233] Step a: Extract relevant data and input it into the causal reasoning module of the digital twin coupling. By analyzing the coupling relationship of multiple factors, infer hidden dangers, trace the root causes, and predict future risk changes.

[0234] Specifically, firstly, geometric and mechanical property data of structural components, such as beam dimensions and material strength, are extracted from the BIM+GIS model. Simultaneously, information on changes in the construction site environment, such as dust concentration and temperature variations, is collected from drone imagery streams. Furthermore, real-time monitoring data, such as crane load and operating frequency, is collected from IoT devices. This data is then preprocessed, including normalization and noise reduction, to meet the needs of subsequent analysis.

[0235] Next, the preprocessed data is input into the causal reasoning module, where algorithms such as Granger causality analysis are used to identify potential causal relationships within the data. For example, if the data shows that crane failure rates increase when workers are working under the crane, this analysis will help confirm whether this activity is a direct cause of the failures.

[0236] Machine learning algorithms such as random forests or gradient boosting trees can be used to analyze the interaction between environmental parameters and equipment performance, thereby identifying potential hidden problems. For example, the analysis may reveal an increased equipment failure rate under high-temperature conditions, which could be a previously undetected problem.

[0237] Finally, time series analysis methods such as the ARIMA model are used to predict future risk trends, such as construction delays that may be caused by low winter temperatures. Based on the analysis results, potential hidden dangers are deduced, such as structural damage caused by equipment overheating, or increased accident risks due to specific construction activities. This information will be integrated into the final report to provide decision support for construction safety management.

[0238] Step b, prior to the report generation step, also includes an emergency response step: based on the risk level of the hidden danger, the cause, and the prediction of future risks, the emergency resource twin coupling module is activated, and a timely and resource-adaptive disposal path is generated through a multi-objective optimization algorithm, and the disposal path is integrated into the final report.

[0239] First, step b employs risk matrix analysis to quantitatively assess potential hazards. This technique assigns risk values ​​by considering the severity and probability of occurrence of the hazard. For example, if the analysis shows that the support column is at high risk due to material fatigue, then this hazard will be assigned a higher risk score.

[0240] Secondly, time series analysis techniques, such as the ARIMA model, can be used to predict the trend of potential hazards over time, thereby anticipating future risks. By analyzing historical construction data and environmental monitoring data, it is possible to predict that the risk of construction delays due to climate change may increase in the coming months.

[0241] Next, the emergency resource twin coupling module is activated. This module uses resource optimization algorithms, such as linear programming, to determine how to most effectively allocate emergency resources on site. For example, in the case of high-risk support pillars, the algorithm might suggest prioritizing the allocation of reinforcement materials and specialized engineering resources to that area.

[0242] Furthermore, multi-objective optimization algorithms, such as genetic algorithms, can be used to generate a disposal path that considers both timeliness and resource suitability. These algorithms can find the optimal balance among multiple objectives, such as minimizing risk, maximizing resource utilization efficiency, and minimizing cost.

[0243] Finally, the generated treatment paths are integrated into the final report, and data visualization techniques, such as GIS mapping, are used to clearly demonstrate the recommended measures and expected effects.

[0244] The multimodal data-driven method for identifying safety hazards in rail transit construction also includes a closed-loop hazard mitigation process and a model self-learning phase, specifically including:

[0245] Step A: After receiving a report containing hazard information, the construction management terminal calls upon a pre-set hazard and mitigation measure association library based on the explicit hazard type (e.g., personnel entering a danger zone, equipment displacement exceeding tolerance) and implicit risk level in the report. This association library matches corresponding mitigation plans according to the three dimensions of hazard type, implicit risk level, and work site characteristics (e.g., for explicit hazards such as segment displacement combined with implicit high-risk stress exceeding limits, a combination of adjusting shield thrust parameters and local grouting reinforcement is matched). The corresponding hazard component is located using the element number in the report, and a targeted mitigation instruction is generated. Simultaneously, a pre-set mitigation ledger generation algorithm is activated to record the mitigation personnel ID, mitigation measure details, mitigation start and completion time, and preliminary results. Then, the ledger information is bound to the BIM+GIS model attribute column of the corresponding spatial element through the element number, realizing the three-dimensional association of hazard, mitigation, and component.

[0246] Step B: After the handling is completed, according to the preset data collection rules for re-inspection (collect 3 sets of images at different time periods for high-risk hazards, and collect 1 set of images for medium and low-risk hazards), the re-inspection image stream of the hazard area is obtained through the drone, and at the same time, real-time monitoring data (such as component stress, ambient temperature and humidity, and equipment operating parameters) from the IoT terminal is retrieved.

[0247] The image feature extraction algorithm used in the visible hazard identification stage is adopted to extract dynamic and static features from the review images, compare them with the hazard features before treatment, and calculate the degree of elimination of visible hazards. When the degree of elimination reaches 90% or above, it is determined that the visible hazard has been eliminated (the specific threshold can be flexibly adjusted according to the importance of the work site).

[0248] The mechanical calculation algorithm and environmental erosion algorithm used in the hidden danger simulation stage are invoked. The re-inspection monitoring data are substituted into the calculation and compared with the quantitative value of hidden risks before treatment. The reduction of hidden risks is calculated. If the reduction reaches 80% or more, the hidden risks are judged to be controllable (the specific threshold can be flexibly adjusted according to the type of component).

[0249] If both the elimination of visible hazards and the reduction of hidden risks meet the standards, the hazard handling is deemed complete; if not, the preset instruction optimization rules are invoked (e.g., if the reduction of hidden risks is insufficient, the frequency of component stress monitoring is increased and the parameters of reinforcement measures are adjusted), and the process returns to step A to reissue the optimization handling instruction, forming a complete closed loop of identification, push, handling, and verification.

[0250] Step C: Activate the preset effective sample screening algorithm. For hazard instances where the closed-loop treatment has been completed, select effective samples according to a preset threshold of ≥0.85: (explicit hazard elimination rate × 0.4 + implicit risk reduction rate × 0.6). (This weight can be dynamically adjusted according to the safety requirements of the work site). Extract all data of the effective samples, including multimodal raw data before treatment (image features, personnel positioning data, equipment and environmental monitoring data), treatment process data (measure details, ledger records), and post-treatment verification data. Then, label the samples according to the hazard type and treatment effect, and assign sample weights (high-risk sample weight is set to 1.2, and medium-risk sample weight is set to 1.0).

[0251] Regularly optimize the explicit hazard identification algorithm based on newly added valid samples: expand the training set by pre-set feature enhancement algorithms (such as adding rotation and noise disturbance processing to abnormal samples of small-sized equipment), and adjust the anchor box parameters and confidence threshold of the target detection model (such as adjusting the detection confidence threshold of personnel entering the danger zone from 0.7 to 0.68 to improve the recognition rate of small-scale intrusion scenarios).

[0252] Optimize the coupling model of hidden dangers: adopt a preset weight optimization algorithm (such as particle swarm optimization algorithm), and iteratively update the preset weighted fusion rules of the model based on the correlation between the treatment effect and the parameters in the sample (such as optimizing the basic weight of environmental indicators of water-rich strata work sites from 0.3 to 0.32, and fine-tuning the environmental correction coefficient of the lining and maintenance stage from 1.15 to 1.18).

[0253] After optimization, the optimization effect is verified by using a preset model performance evaluation algorithm with explicit identification accuracy and implicit risk value error rate as the core indicators. If the target is met, the system algorithm model is updated to achieve continuous evolution of the hidden danger identification capability; if the target is not met, the sample screening conditions and optimization parameters are reviewed and the optimization process is re-executed.

[0254] Based on the same inventive concept, embodiments of the present invention provide a multimodal data-driven system for identifying safety hazards in rail transit construction, including a memory and a processor. The memory stores data that can run on the processor to implement the following... Figure 1 The procedure for the method shown.

[0255] The embodiments described in this specific implementation are preferred embodiments of this application and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A multimodal data-driven method for identifying safety hazards in rail transit construction, characterized in that, include: Acquire multimodal data of the target rail transit project, including BIM+GIS models of geometric and mechanical properties of components at construction sites, continuous UAV image streams, personnel positioning data from IoT terminals, equipment operating parameters, and environmental monitoring data; Based on the preset edge-cloud collaborative architecture, data is fused with the spatial coordinate system of the BIM+GIS model as the reference: After receiving the continuous image stream from the drone, personnel positioning data, equipment operating parameters, and environmental monitoring data, the edge device performs differentiated spatial association according to preset rules, completes the binding of the four types of data with the spatial elements of the model, and transmits them to the cloud after cleaning and compression. The cloud calls up all attributes of the BIM+GIS model, corrects correlation deviations, integrates the data with the geometric and mechanical attributes and element numbers of the corresponding spatial elements, and generates a standardized dataset containing element numbers, spatial coordinates, time series, element attributes and monitoring data. Perform hierarchical identification by calling a standardized dataset: use a preset algorithm to extract image stream features, combine personnel positioning data to identify visible hazards and mark them to corresponding spatial elements, and associate element numbers; for the marked elements and associated element numbers, extract relevant data and input them into a preset coupling model to deduce hidden hazards and output element risk values, which are mapped to risk levels; The results of the two-level correlation verification are used to match spatial elements with their corresponding risk levels based on element numbers. The comprehensive risk index of each element number is calculated using the entropy weight method, and a report containing the hazard type, element number, comprehensive risk index, risk level and data source is generated and pushed to the construction management terminal.

2. The method for identifying safety hazards in rail transit construction driven by multimodal data according to claim 1, characterized in that, Perform differentiated spatial association according to preset rules to complete the binding of four types of data with model spatial elements, including: For continuous image streams, a preset feature extraction algorithm is used to extract dynamic target and static area feature points. Combined with BIM+GIS model parameters, bundle adjustment is performed to establish a unified spatial benchmark with model spatial elements as anchor points to achieve dynamic coordinate matching. Based on the above benchmarks, a preset filtering algorithm is introduced to optimize the trajectory of personnel positioning data, anchoring it to the corresponding elements, associating personnel type attributes, and comparing it with the spatiotemporal targets of personnel in the image to form double verification. For equipment operating parameters, parse the equipment ID and call the preset BIM equipment-work surface binding relationship to associate it with the current work element; Based on the above benchmarks, the coverage of environmental monitoring data is expanded using a preset interpolation algorithm, and an element-level environmental parameter field is generated by combining preset topological relationships. Using a pre-defined multi-source data correlation evaluation model, the spatial consistency and temporal correlation of related data for the same element are calculated, data with deviations exceeding the threshold are removed, and the consistency of the correlation is verified. Based on the verification results, a spatiotemporal correlation map with model spatial elements as the core is generated, realizing the three-dimensional binding of spatial coordinates, entity attributes and security semantics.

3. The method for identifying safety hazards in rail transit construction driven by multimodal data according to claim 2, characterized in that, Based on the verification results, a spatiotemporal correlation map centered on model spatial elements is generated, realizing the three-dimensional binding of spatial coordinates, entity attributes, and security semantics, including: The map is initialized with consistent model spatial features as core nodes. The nodes integrate the spatial coordinates corresponding to the unified spatial benchmark and the geometric and mechanical attributes in the BIM+GIS model parameters. Spatial adjacent edges between features are generated based on spatial coordinates, and mechanically dependent edges are generated based on geometric and mechanical attributes to construct a network of relationships. Based on the coordinate association of spatially adjacent edges and the attribute association of mechanically dependent edges, the continuous image stream feature points, optimized personnel positioning trajectories, equipment operating parameters and element-level environmental parameter fields that have passed the association degree evaluation are transformed into dynamic data blocks in time sequence. These blocks are then associated with the corresponding core nodes through data attachment edges, and the attribute records of these data attachment edges record the data acquisition timestamp and source reliability. By calling the preset safety rule library and combining the attributes of personnel types and the binding relationship between BIM equipment and work surfaces, dynamic data is matched with semantic templates to generate real-time safety status labels; Based on the correlation parameters of spatially adjacent edges and mechanically dependent edges, the scope of environmental anomaly impact is calculated using a pre-set diffusion model for elements connected by spatially adjacent edges, and the parameter chain effect is analyzed using a pre-set structural mechanics model for elements connected by mechanically dependent edges, and the derived risk semantics are labeled. Establish a dynamic update mechanism: After receiving new verification data, locate the corresponding core node, synchronously update the dynamic data associated with its spatial coordinates and supplementary information of entity attributes, regenerate the real-time security status label and update the derived risk semantics, and maintain the real-time three-dimensional binding of spatial coordinates, entity attributes and security semantics.

4. The method for identifying safety hazards in rail transit construction driven by multimodal data according to claim 3, characterized in that, The semantics of derived risk annotation include: Extract the correlation parameters of spatially adjacent edges and the correlation parameters of mechanically dependent edges; For elements connected by spatially adjacent edges, abnormal data exceeding the safety threshold in the element-level environmental parameter field are input into a preset diffusion model. Combined with the spacing parameter of the spatially adjacent edges, the diffusion range, concentration decay coefficient, and time-effect curve of environmental anomalies among related elements are calculated, and the boundary of environmental anomaly impact is output. For elements connected by mechanically dependent edges, extract the geometric and mechanical properties of the core nodes and the equipment operating parameters in the dynamic data block, input the preset structural mechanics model, combine the load transfer coefficient of the mechanically dependent edge, analyze the chain effect caused by parameter anomalies, and output the mechanical response intensity. Based on the boundary of environmental anomalies output by the diffusion model and the mechanical response intensity output by the structural mechanics model, a preliminary version of the derived risk semantics for related elements is generated by matching the pre-set engineering risk semantic library. The initial semantics were revised by combining real-time security status tags to clarify the risk causes and affected objects, thus forming the final derived risk semantics.

5. The method for identifying safety hazards in rail transit construction driven by multimodal data according to claim 3, characterized in that, Accessing all attributes of the BIM+GIS model from the cloud and correcting correlation discrepancies includes: The cloud accesses all attributes of the BIM+GIS model and combines them with the unified spatial benchmark established at the edge to build a benchmark framework for correcting related deviations. Based on the above benchmark framework, the correlation deviation between feature points in continuous image streams and spatial elements in the model is addressed: the RANSAC algorithm is used to remove mismatched feature points, and then perspective transformation is performed in conjunction with the geometric attributes of the BIM+GIS model to achieve accurate alignment between the spatial coordinates of image feature points and corresponding spatial elements in the model. To address the discrepancy between personnel data and spatial features in the model, a dynamic time warping algorithm is introduced to align temporal relationships based on the optimized personnel location trajectories at the edge. Simultaneously, the work area zoning attributes of the BIM+GIS model are combined to correct spatiotemporal misalignments, thus forming a precise association between personnel trajectories and features. To address the discrepancy between the correlation between equipment operating parameters and model space elements, based on the preset BIM equipment-work surface binding relationship, the mapping relationship between equipment ID and work elements is analyzed, and attribute correlation discrepancies are corrected through parameter matching algorithms to ensure the consistency of the correlation between equipment operating parameters and corresponding elements. To address the correlation deviation between the element-level environmental parameter field and the spatial elements of the model, a spatial interpolation algorithm is used to optimize the spatial distribution of the parameter field based on the spatial topological relationship of the BIM+GIS model in the benchmark framework. Combined with the preset topological relationship, the correlation deviation between environmental data and the spatial range of elements is corrected, so that the environmental parameter field accurately covers the corresponding elements.

6. The method for identifying safety hazards in rail transit construction driven by multimodal data according to claim 5, characterized in that, By combining preset topological relationships to correct the correlation deviation between environmental data and the spatial extent of elements, the environmental parameter field accurately covers the corresponding elements, including: Extract the spatial topological relationships of the BIM+GIS model in the benchmark framework, including the inclusion, adjacency, and subordination relationships between elements, and construct a topological rule library for element-environment association by combining it with preset engineering topological rules; The three-dimensional coverage of the optimized element-level environmental parameter field is analyzed, and the geometric boundaries of the corresponding spatial elements are extracted from the BIM+GIS model to generate a comparison matrix between the parameter field boundary and the element boundary. Based on the comparison matrix, identify and process the type of deviation: for redundant areas in the parameter field that exceed the feature boundary, trim them according to the feature hierarchy in the topology rule base; for feature areas not covered by the parameter field, extract data from neighboring parameter fields based on the relationship between adjacent features, and supplement the coverage through weighted interpolation. A spatial fit calculation model is introduced to quantify the matching degree between the corrected parameter field and the spatial range of the elements. If the matching degree is lower than the preset threshold, return to the previous step to re-optimize the topological association rules and interpolation parameters; if the matching degree is higher than or equal to the threshold, proceed to the next step. By associating the timestamps of dynamic data blocks, consistency verification is performed on the parameter field coverage of the same element at different time series, correcting the time series correlation deviation caused by dynamic changes in construction, and ensuring the coverage accuracy of the spatiotemporal dimension. Output an element-level environmental parameter field whose coverage area precisely matches the spatial boundary of the element, and record the basis for correction.

7. The method for identifying safety hazards in rail transit construction driven by multimodal data according to claim 1, characterized in that, Image stream features are extracted using a pre-defined algorithm, and visible hazards are identified and annotated to corresponding spatial elements based on personnel location data, including: Extract the basic data associated with the labeled features, including continuous UAV image streams, real-time personnel positioning data, and BIM+GIS coordinate information of the corresponding model spatial features; The continuous image stream of the UAV is processed by a preset image feature extraction algorithm. First, redundant frames are removed by the inter-frame difference method, and then the target is identified by the preset target detection model. The pixel coordinates and category labels of each target are output. The real-time positioning data of personnel is preprocessed using a preset positioning optimization algorithm to optimize the original trajectory to eliminate signal drift and output the real-time spatial coordinates of personnel that are consistent with the BIM+GIS model coordinate system. The target information extracted from the image is associated with the model spatial elements using a preset coordinate mapping method. The image target pixel coordinates are converted into model absolute coordinates, and the preset spatial element type is matched with the coordinate range. The spatial correlation results are combined with the preset explicit hazard rule base for judgment. If the personnel coordinates trigger the danger zone range, the equipment coordinates exceed the preset position deviation threshold, or the component features match the damage label, it is judged as an explicit hazard. Identified visible hazards are labeled to the corresponding model space elements, and the hazard type, timestamp, and target coordinates are recorded.

8. The method for identifying safety hazards in rail transit construction driven by multimodal data according to claim 7, characterized in that: For the labeled elements and related element numbers, relevant data is extracted and input into a preset coupling model to deduce hidden risks and output element risk values, which are mapped to risk levels, including: Based on the element number, the associated data of the labeled elements are extracted, including explicit hazard labeling information, geometric and mechanical properties of the BIM+GIS model, equipment operating parameters, and element-level environmental monitoring data; The extracted associated data is processed using a pre-defined standardized algorithm: the units and time granularity of equipment operating parameters are unified, missing time-series values ​​in environmental monitoring data are filled, explicit hazard labeling information is converted into quantitative triggering factors, and a standardized multi-dimensional parameter set is output. The standardized multi-dimensional parameter set is associated and mapped with the geometric and mechanical properties of the BIM+GIS model: the equipment operation parameters are bound to the mechanical parameters of the corresponding work surface, and the environmental monitoring data is associated with the material properties of the components to generate an element-parameter association matrix. The element-parameter correlation matrix is ​​input into the preset coupling model, and the preset fusion algorithm is used for hierarchical quantification: the cumulative deformation and stress over-limit frequency are calculated by combining the mechanical parameters of the component and the operating parameters of the equipment through the preset mechanical calculation algorithm; the corrosion depth and durability decay rate are inferred by associating material properties with environmental monitoring data through the preset environmental erosion algorithm; and the hidden danger comprehensive quantification value is output according to the preset weighted fusion rules, which serves as the element risk value. The risk level of the element is determined by matching the comprehensive quantitative value with the corresponding level using a preset risk level mapping rule.

9. A multimodal data-driven system for identifying safety hazards in rail transit construction, characterized in that, The method includes a memory, a processor, and a program stored in the memory and executable on the processor, which, when loaded and executed by the processor, implements a multimodal data-driven method for identifying safety hazards in rail transit construction as described in any one of claims 1 to 8.

Images