Building construction site safety risk real-time early warning system based on digital twinning

Abstract

The invention discloses a building construction site safety risk real-time early warning system based on digital twinning. The building construction site safety risk real-time early warning system comprises a data acquisition layer, a data fusion processing layer, a digital twinning dynamic mapping layer, a safety risk identification and early warning layer and an early warning output and response layer. A technical path for extracting risk characteristics from a digital mirror image by a digital twinning dynamic mapping layer is defined, a data fusion model combining an improved D-S evidence theory and a long-short-term memory network model is established, and deep interaction between a twinning model and a risk assessment model is realized. According to the structure, a risk feature extraction mechanism of the digital twinning dynamic mapping layer can be defined, and real-time identification and efficient disposal of safety risks are realized.

Description

Technical Field

[0001] This invention relates to the field of building construction technology, specifically a real-time early warning system for safety risks at building construction sites based on digital twins. Background Technology

[0002] Construction sites are characterized by complex environments, frequent personnel movement, and diverse equipment types. Safety risks are characterized by suddenness, coupling, and dynamism. Traditional safety management relies on manual inspections and fixed threshold monitoring, which suffers from problems such as delayed risk identification, high false alarm and missed alarm rates, and fragmented data.

[0003] While existing digital twin-based safety management systems have achieved digital mapping of construction sites, they still have key technical shortcomings: 1. The ability to fuse multi-source heterogeneous data is weak, and it is unable to effectively integrate data from sensors, videos, positioning, etc., resulting in insufficient data value mining; 2. The digital twin model update is disconnected from risk feature extraction. The model is only used as a visualization carrier, and the technical path for obtaining risk features from the digital image is not clearly defined. 3. Insufficient interaction between the risk assessment model and the twin model, and the lack of a linkage mechanism between data, models, and assessments; 4. The early warning response lacks dynamic adaptation with the twin model, and the feasibility of the disposal plan cannot be verified by the model.

[0004] Therefore, there is an urgent need to develop a real-time early warning system for safety risks at construction sites based on digital twins to solve the problems in existing technologies. Summary of the Invention

[0005] The purpose of this invention is to provide a real-time early warning system for safety risks at construction sites based on digital twins. This system can clearly define the risk feature extraction mechanism of the dynamic mapping layer of the digital twin, realize the real-time identification and efficient handling of safety risks, and has a simple structure and is easy to use, thereby solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: A real-time early warning system for safety risks at construction sites based on digital twins includes a data acquisition layer, a data fusion and processing layer, a digital twin dynamic mapping layer, a safety risk identification and early warning layer, and an early warning output and response layer. The data acquisition layer serves as the system's data input terminal, collecting multi-source heterogeneous data from the construction site. This multi-source heterogeneous data includes physical entity data, environmental data, behavioral data, and business data, and outputs the collected data to the data fusion processing layer in real time. The data fusion processing layer cleans, standardizes, and fuses the multi-source heterogeneous data. It uses an improved DS evidence theory combined with a long short-term memory network model to eliminate data uncertainty and achieve spatiotemporal alignment. The processed unified feature data is then output to the digital twin dynamic mapping layer. The digital twin dynamic mapping layer is based on building information modeling and geographic information system (GIS) to build a digital twin model. The digital twin model includes three dimensions: geometric, physical and behavioral. The model status is updated in real time based on the input feature data, forming a digital mirror of the construction site and outputting risk feature data to the safety risk identification and early warning layer. The security risk identification and early warning layer identifies entity risks using an improved YOLOv8 model, assesses risk levels using an analytic hierarchy process-fuzzy comprehensive evaluation model, generates graded early warning signals based on the assessment results, and outputs them to the early warning output and response layer. The early warning output and response layer outputs early warning information and generates handling plans through multiple terminals, while collecting risk handling results and feeding them back to the digital twin dynamic mapping layer to complete the closed-loop update.

[0007] Preferably, the data acquisition layer collects multi-source heterogeneous data from the construction site, including the following modules: IoT sensor module: Collects physical entity data and environmental data, including force sensors and stress sensors to collect equipment operating parameters, and temperature, humidity, dust and wind speed sensors to collect environmental parameters, and outputs numerical data at a preset frequency; Video acquisition module: includes camera equipment to capture images of the construction site and output image data for identifying personnel behavior and equipment status; Positioning module: Collects the location of personnel and mobile devices using ultra-wideband positioning tags and outputs location data; Business Interface Module: Connects to the construction management system through standardized interfaces and outputs corresponding business data.

[0008] Preferably, the standardized output of the data fusion layer includes the following: It receives multi-source heterogeneous data from the data acquisition layer, performs standardization processing on numerical data, encoding processing on text data, and frame extraction on image data, removes abnormal data and fills in missing data, and outputs standardized data. Standardized data is mapped to a unified spatiotemporal coordinate system, and temporal features are processed through a long short-term memory network model to output a time-aligned multidimensional feature sequence. By introducing an evidence weighting coefficient, the DS evidence theory is improved, the problem of data conflict is resolved, and credible fused data is output.

[0009] Preferably, the risk feature data output by the digital twin dynamic mapping layer specifically includes the following: Building element models are constructed using Building Information Modeling (BIM), and terrain and safety zone models are constructed using Geographic Information System (GIS). These models are then integrated into a unified basic model through coordinate transformation, and a twin base containing full element information is output. The unified feature data output by the data fusion processing layer is used to trigger changes in position and parameters through events and is combined with timed inspections to synchronously update the geometric position, physical properties and behavioral state of the model, and output a real-time twin image. Static features, including the extent of hazardous areas and equipment rated parameters, are extracted from the real-time twin image, while dynamic features, including personnel locations and equipment operating parameters, are output as a standardized risk feature vector.

[0010] Preferably, the digital twin dynamic mapping layer also includes a lightweight rendering module, which receives the real-time updated twin model, performs lightweight processing, and outputs an interactive rendering model through a 3D engine.

[0011] Preferably, the security risk identification and early warning layer assesses the risk level and generates graded early warning signals based on the assessment results, including the following: The system receives the risk feature vector output from the digital twin dynamic mapping layer, optimizes the anchor box YOLOv8 model through K-means clustering, and identifies risk targets such as personnel violations and equipment malfunctions by combining the attention mechanism, and outputs the entity risk results. Based on the entity risk results, the weights of the indicators are determined by the analytic hierarchy process (AHP) algorithm, the membership degree is determined by fuzzy numbers, and the quantitative risk level is calculated and output by the fuzzy hierarchical comprehensive evaluation model. Receive the quantified risk level, trigger the early warning according to the preset rules, and output the risk-related information simultaneously, including the risk location and risk type.

[0012] Preferably, the early warning output and response layer generating a handling plan and completing a closed-loop update includes the following: Receive early warning signals and output them synchronously through the output device; Receive risk type and level information, call the preset handling solution library through the rule engine, and output differentiated handling strategies; The results of the actions, whether manually entered or automatically monitored, are fed back to the digital twin dynamic mapping layer to update the model status and store the action records.

[0013] Preferably, in the early warning output and response layer, the preset solution library built into the rule engine contains the correspondence between risk types and disposal measures, and the solution library is updated according to changes in the construction scenario. Before the disposal strategy is output, it is sent to the digital twin model and its feasibility is verified through simulation. If it is feasible, it is output; if it is not feasible, it is terminated.

[0014] Preferably, an edge-cloud collaboration module is also included, serving as a support layer for data transmission and storage to achieve layered processing. This includes edge gateways, which are deployed at construction sites to receive multi-source heterogeneous data output from the data acquisition layer, perform local simplification processing, and output the simplified data to the cloud, reducing transmission pressure. This includes cloud nodes, which use elastic servers to receive simplified data and operational logs from the edge gateway, store twin models, handling records, and other data, and output storage services to support interactions between different layers.

[0015] This application also discloses a computer-readable storage medium having a program stored thereon, which, when executed by a processor, implements the aforementioned real-time early warning system for safety risks at construction sites based on digital twins.

[0016] Compared with the prior art, the beneficial effects of the present invention are: By improving the DS evidence theory and combining it with the LSTM network fusion processing structure, the spatiotemporal alignment and conflict elimination of multi-source heterogeneous data were achieved, which effectively improved the accuracy of fused data and greatly improved the data conflict resolution rate compared with traditional Kalman filtering. This significantly improved the accuracy and reliability of data fusion. By improving the early warning core structure of YOLOv8 entity recognition combined with AHP-FCE level assessment, accurate identification and quantitative classification of construction risks have been achieved. With the improved YOLOv8's extremely high recognition accuracy and the AHP-FCE model's extremely low false alarm rate, it is superior to the existing threshold judgment method, which has the effect of improving the accuracy of risk identification and reducing invalid early warnings. Through a closed-loop response structure that outputs early warnings and provides feedback on the results of solution generation, efficient management of safety risks from discovery to handling has been achieved. The average time for risk handling has been further shortened, efficiency has been further improved, safety management efficiency has been significantly enhanced, and construction safety risks have been reduced.

[0017] Other features and advantages of the present invention will be disclosed in detail in the following detailed description and accompanying drawings. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structure of a real-time early warning system for safety risks at a construction site based on digital twins, as described in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] In this embodiment of the invention, a real-time early warning system for safety risks at construction sites based on digital twins is described below. Figure 1 As shown, it includes a data acquisition layer, a data fusion and processing layer, a digital twin dynamic mapping layer, a security risk identification and early warning layer, and an early warning output and response layer. Each layer is connected sequentially according to the logic of data input-processing-output. At the same time, the early warning output and response layer feeds back data to the digital twin dynamic mapping layer, forming a closed loop. 1) Data Acquisition Layer: Acquires heterogeneous data from multiple sources and outputs it to the fusion processing layer; 2) Data fusion processing layer: processes the collected data and outputs unified feature data to the twin mapping layer; 3) Digital Twin Dynamic Mapping Layer: Updates the twin model based on feature data, extracts risk features, and outputs them to the risk warning layer; 4) Safety Risk Identification and Early Warning Layer: Identifies risks and outputs early warning signals to the output response layer; 5) Early warning output and response layer: Outputs early warning and response plans, and collects and feeds back the response results to the twin mapping layer.

[0021] (1) Data acquisition layer: multi-dimensional data input mechanism As the system's data input end, a multi-source acquisition method is adopted, combining IoT sensors with video acquisition, positioning modules, and business interfaces. The output ends of each module are all connected to the data fusion and processing layer to ensure data comprehensiveness and real-time performance. Module type Core equipment / interfaces Content collected Output format Transmission Protocol IoT sensor module Force sensors, stress sensors, temperature and humidity sensors, dust sensors, wind speed sensors Equipment operating parameters (torque, load, stress), environmental parameters (temperature, humidity, PM2.5, wind speed) Numerical data stream (preset frequency 1-5Hz) MQTT (QoS=1) Video capture module High-definition camera with 2 megapixels or higher (supports infrared) Personnel operation behavior, equipment operating status, and safety sign recognition Image frame stream (25fps) RTSP Positioning module UWB positioning tag (positioning accuracy ±0.3m) Real-time location of personnel and mobile devices Coordinate data (10Hz refresh rate) UWB proprietary protocol Business Interface Module Standardized RESTful API interface Construction procedures, personnel qualifications, equipment maintenance records Structured text data (30 min / time) HTTP / HTTPS The data collected by each module is transmitted through a hybrid wired and wireless transmission method. Fixed devices such as temperature and humidity sensors use wired connections to ensure stability, while mobile devices such as UWB tags use wireless transmission to improve flexibility, ensuring that the data is output to the data fusion processing layer in real time.

[0022] (2) Data fusion processing layer: standardized data output mechanism The system receives raw data from the data acquisition layer, processes it according to a preprocessing, spatiotemporal fusion, and uncertainty optimization process, and outputs unified feature data to the digital twin dynamic mapping layer. The core technology is the combination of improved DS evidence theory and LSTM network. The preprocessing submodule employs differentiated processing methods for different data types: numerical data, such as sensor parameters, is standardized using Z-Score, mapping the data to the 0-1 range; text data, such as personnel qualifications, is converted to vectors using One-Hot encoding; image data undergoes frame extraction and grayscale conversion, removing blurry frames. Outlier data, such as sudden sensor values, is removed using the 3σ criterion, and missing data is filled in using linear interpolation. In one feasible implementation, the standard missing rate is ≤5%, outputting standardized data.

[0023] The spatiotemporal fusion submodule constructs a unified spatiotemporal coordinate system, using the GPS point at the southwest corner of the construction site as the origin and synchronizing the time to UTC. Standardized data with different timestamps and locations are mapped to this coordinate system. An LSTM network is used to process temporal features; the network structure is input layer → hidden layer → Dropout layer → output layer, outputting a time-aligned multidimensional feature sequence.

[0024] In one feasible embodiment, the input layer is 48-dimensional, corresponding to 8 types of devices × 6 features → the hidden layer is 128-dimensional, with ReLU activation → the dropout layer has a dropout rate of 0.2 → the output layer is 48-dimensional. Uncertainty Optimization Submodule: This module introduces evidence weight coefficients to improve traditional DS evidence theory and resolve data conflict issues. It calculates the similarity between pieces of evidence using the Jousselme distance formula to determine weight allocation. The sum of these weights is 1, and these weights are integrated into the evidence combination rules to enhance the credibility of the fusion result. For example, when fusing three pieces of evidence, if evidence 1 has high similarity to other evidence, it is assigned a higher weight, ultimately outputting highly credible unified feature data.

[0025] (3) Digital Twin Dynamic Mapping Layer: Core Mechanism for Data Feature Transformation Receiving unified feature data from the data fusion processing layer, constructing a twin model based on BIM and GIS, enabling linked model updates and feature extraction, and outputting standardized risk feature vectors to the safety risk identification and early warning layer are the core steps in the system's data early warning conversion. In one feasible embodiment, the technical details are as follows: Basic modeling capabilities: A BIM model was built using Revit 2024 with a LOD of 400, including building components, construction equipment, personnel, and other elements, with attribute information covering dimensions, materials, and rated parameters. A GIS model was built using ArcGIS Pro 3.2, including terrain, surrounding environment, and safety zone delineation. The BIM and GIS models were integrated through a seven-parameter coordinate transformation, stored in a PostgreSQL database, and a full-feature twin base was output.

[0026] Dynamic update capability: Employing a dual-mode update model of "event triggering + scheduled inspection" to ensure synchronization with the physical site: ① Geometric update: When the equipment position changes by ≥0.5m or the component installation is completed, the model coordinates are updated using linear interpolation algorithm based on UWB positioning data and BIM parameters, with an update delay of ≤100ms; ② Physical update: When equipment / environmental parameters exceed the rated value by ±5%, the physical properties of the model are updated synchronously. For example, when the tower crane torque is ≥90% of the rated value, the tower crane icon in the model will be displayed in red. ③ Behavior Update: When personnel enter a danger zone, such as when they are ≤1m from the boundary, or when the equipment status changes, the system simulates personnel walking, equipment operation, and other behaviors, with the personnel icon flashing at a frequency of 2Hz. The updated model forms a real-time digital mirror.

[0027] Risk feature extraction capability: Extracting both static and dynamic risk features from real-time digital images to form standardized risk feature vectors; the extraction logic and technical path are clearly defined. ① Static feature extraction: Read preset parameters from the BIM model attribute library, including the range of hazardous areas, equipment rated parameters, and personnel qualification levels, and quickly call them through SQL query statements. The extraction frequency is synchronized with business data. ② Dynamic feature extraction: Based on real-time mirroring driven by fused data, the real-time location of personnel, equipment operating parameters, environmental parameters, and construction process progress are extracted through coordinate comparison, and the extraction frequency is synchronized with the fused data; ③ Feature standardization: Static and dynamic features are uniformly mapped to the 0-1 interval to construct a standardized risk feature vector with the dimension of the number of risk types × the number of features, which is then output to the safety risk identification and early warning layer.

[0028] Lightweight rendering module: It uses WebGL technology to lightweight process the fused model and implements browser rendering through the Three.js engine. It supports interactive operations such as scaling, rotation, and slicing, ensuring smooth display on mobile and PC, and providing a visual operation interface for administrators.

[0029] (4) Safety Risk Identification and Early Warning Layer: Feature-Based Early Warning Precision Conversion Mechanism It receives risk feature vectors output from the digital twin dynamic mapping layer, operates according to the process of entity identification, level assessment, and early warning triggering, and outputs graded early warning signals to the early warning output and response layer. The core technology is the combination of an improved YOLOv8 model and an AHP-FCE model. Entity recognition unit: Inputting image and location features from the risk feature vector, it identifies risk targets using an improved YOLOv8 model. Anchor box size is optimized through K-means clustering, and a CBAM attention mechanism is introduced into the model neck to enhance small target recognition capabilities.

[0030] The training dataset consists of 10,000 labeled construction site images. The model achieves 98.2% mAP@0.5 on the test set and can identify entity risks such as personnel not wearing safety helmets and equipment malfunctions, outputting entity risk results including risk type, location, and confidence level.

[0031] Risk assessment unit: Input entity risk results and feature vectors, and use AHP-FCE model to quantitatively assess risk level.

[0032] ① Construct an indicator system: The primary indicators are: personnel risk (weight 0.35), equipment risk (weight 0.30), environmental risk (weight 0.20), and management risk (weight 0.15). The secondary indicators include 12 items such as wearing a safety helmet (weight 0.4) and tower crane torque (weight 0.4). ②AHP weight calculation: The judgment matrix is ​​constructed using the 1-9 scaling method. After verifying consistency, the indicator weights are determined. In one feasible embodiment, the consistency ratio CR < 0.1. ③ Fuzzy evaluation: The membership degree is determined by triangular fuzzy numbers. For example, not wearing a safety helmet corresponds to a high risk membership degree of 0.8, 0.9, and 1.0. ④ Comprehensive Calculation: The evaluation results are obtained through a weighted average operator and mapped to four levels: Level I, low risk, 0-0.25; Level II, moderate risk, 0.25-0.5; Level III, relatively high risk, 0.5-0.75; Level IV, high risk, 0.75-1.0; output quantitative risk level.

[0033] Warning triggering unit: Triggers warnings according to preset rules, with a response time of ≤1s.

[0034] For example: Level III, which is relatively high risk, outputs a local early warning signal, with sound and light alarms at the construction site and management personnel's APP; Level IV, which is high risk, outputs a multi-level linkage early warning signal, which is also synchronized to the company headquarters and the supervision platform, and outputs the risk location, GPS coordinates, twin model positioning, and related information such as type and level.

[0035] (5) Early warning output and response layer: Early warning feedback closed-loop mechanism It receives warning signals from the security risk identification and early warning layer, realizes a closed loop of multi-terminal output, solution generation, and result feedback, and interfaces with the digital twin dynamic mapping layer at the feedback end. Multi-terminal output capability: Risk information can be displayed on LED screens, and warning information can be simultaneously output via sound and light alarms, mobile terminal APP push and voice broadcast, and PC platform twin model positioning jump to ensure rapid response of management personnel.

[0036] Solution generation capability: Based on risk type and level, the system calls the preset disposal solution library through the Drools rule engine. For example, the solution for tower crane overload (Level IV) is: remote shutdown → notify the operator → on-site verification. After the solution is generated, the disposal process is simulated through a twin model, such as simulating the tower crane status after shutdown. After verifying feasibility, a differentiated disposal strategy is output.

[0037] Results feedback capability: The system obtains handling results through manual input via APP or automatic monitoring by sensors. For example, if the load is adjusted to a safe range, the system outputs feedback data to the digital twin dynamic mapping layer to update the model status, such as restoring the tower crane icon to blue. The system also stores handling records, including time, personnel, and measures, to complete closed-loop management.

[0038] (6) Edge-Cloud Collaboration Module: Supporting Layer Technology As a system support layer, it enables layered data processing and storage: ① Edge gateway, deployed at the construction site, receives data from the acquisition layer and preprocesses it locally (such as removing invalid frames), and outputs simplified data to the cloud to reduce transmission pressure; ② Cloud nodes store twin models, handling records, and operation logs, support remote access from multiple terminals, and provide data storage and computing services to support interactions at all levels.

[0039] This invention also provides a computer-readable storage medium storing a program thereon, which, when executed by a processor, implements the real-time early warning system for safety risks at construction sites based on digital twins as described in the above embodiments.

[0040] The computer-readable storage medium can be an internal storage unit of any data processing device described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium can also be an external storage device of any data processing device, such as a plug-in hard disk, SmartMediaCard (SMC), SD card, or FlashCard equipped on the device. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices of any data processing device. The computer-readable storage medium is used to store the computer program and other programs and data required by the data processing device, and can also be used to temporarily store data that has been output or will be output.

[0041] Example 1 1. Hardware Deployment (1) Deployment of data acquisition equipment: Torque sensors and wind speed sensors are installed on the top of the tower crane; Weight sensors are installed at the four corners inside the elevator. Stress sensors are attached to the scaffolding every 10 meters. Temperature and humidity sensors and dust sensors are installed every 50m on site; Cameras were installed at entrances and exits; Personnel wear location tags at all times.

[0042] Transmission and early warning equipment: Edge gateways are deployed in power distribution rooms; 5G routers enable cloud-based communication; Install audible and visual alarms at the construction site; Management personnel are equipped with smartphones; The monitoring center is equipped with Intel servers and LED screens.

[0043] 2. Taking personnel illegally entering a dangerous area as an example, the operational process includes: 1. Data Input: UWB tags collect personnel location data, cameras capture images, the business system outputs the danger zone range, and the data is transmitted to the fusion processing layer; 2. Data processing: Preprocessing removes outlier coordinates, LSTM aligns time-series data, improves DS to fuse location and image evidence, and outputs unified feature data; 3. Twin mapping: Feature data drives the model to update personnel location, extract risk features, and output feature vectors when personnel coordinates are within the danger zone and are not wearing safety helmets; 4. Risk Warning: YOLOv8 identifies the absence of a safety helmet, and AHP-FCE calculates a risk value of 0.85, which is Level IV, triggering a multi-level warning; Output response: Multi-terminal alarm, generate: on-site announcement → guide evacuation → safety education plan, feedback results after handling, update personnel position to safe zone in the model, and cancel the warning.

[0044] This invention provides a real-time early warning system for safety risks at construction sites based on digital twins. It can clearly define the risk feature extraction mechanism of the dynamic mapping layer of the digital twin, realize the real-time identification and efficient handling of safety risks, and has high reliability.

[0045] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0046] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A real-time early warning system for safety risks at construction sites based on digital twins, characterized in that: It includes a data acquisition layer, a data fusion and processing layer, a digital twin dynamic mapping layer, a security risk identification and early warning layer, and an early warning output and response layer; The data acquisition layer serves as the system's data input terminal, collecting multi-source heterogeneous data from the construction site. This multi-source heterogeneous data includes physical entity data, environmental data, behavioral data, and business data, and outputs the collected data to the data fusion processing layer in real time. The data fusion processing layer cleans, standardizes, and fuses the multi-source heterogeneous data. It uses an improved DS evidence theory combined with a long short-term memory network model to eliminate data uncertainty and achieve spatiotemporal alignment. The processed unified feature data is then output to the digital twin dynamic mapping layer. The digital twin dynamic mapping layer is based on building information modeling and geographic information system (GIS) to build a digital twin model. The digital twin model includes three dimensions: geometric, physical and behavioral. The model status is updated in real time based on the input feature data, forming a digital mirror of the construction site and outputting risk feature data to the safety risk identification and early warning layer. The security risk identification and early warning layer identifies entity risks through an improved YOLOv8 model, assesses risk levels by combining a fuzzy hierarchical comprehensive evaluation model, generates graded early warning signals based on the assessment results, and outputs them to the early warning output and response layer. The early warning output and response layer outputs early warning information and generates handling plans through multiple terminals, while collecting risk handling results and feeding them back to the digital twin dynamic mapping layer to complete the closed-loop update.

2. The real-time early warning system for safety risks at construction sites based on digital twins as described in claim 1, characterized in that, The data acquisition layer collects multi-source heterogeneous data from the construction site, including the following modules: IoT sensor module: Collects physical entity data and environmental data, including force sensors and stress sensors to collect equipment operating parameters, and temperature, humidity, dust and wind speed sensors to collect environmental parameters, and outputs numerical data at a preset frequency; Video acquisition module: includes camera equipment to capture images of the construction site and output image data for identifying personnel behavior and equipment status; Positioning module: Collects the location of personnel and mobile devices using ultra-wideband positioning tags and outputs location data; Business Interface Module: Connects to the construction management system through standardized interfaces and outputs corresponding business data.

3. The real-time early warning system for safety risks at construction sites based on digital twins as described in claim 1, characterized in that, The standardized output of the data fusion layer includes the following: It receives multi-source heterogeneous data from the data acquisition layer, performs standardization processing on numerical data, encoding processing on text data, and frame extraction on image data, removes abnormal data and fills in missing data, and outputs standardized data. Standardized data is mapped to a unified spatiotemporal coordinate system, and temporal features are processed through a long short-term memory network model to output a time-aligned multidimensional feature sequence. By introducing an evidence weighting coefficient, the DS evidence theory is improved, the problem of data conflict is resolved, and credible fused data is output.

4. The real-time early warning system for safety risks at construction sites based on digital twins as described in claim 1, characterized in that, The risk feature data output by the digital twin dynamic mapping layer specifically includes the following: Building element models are constructed using Building Information Modeling (BIM), and terrain and safety zone models are constructed using Geographic Information System (GIS). These models are then integrated into a unified basic model through coordinate transformation, and a twin base containing full element information is output. The unified feature data output by the data fusion processing layer is used to trigger changes in position and parameters through events and is combined with timed inspections to synchronously update the geometric position, physical properties and behavioral state of the model, and output a real-time twin image. Static features, including the extent of hazardous areas and equipment rated parameters, are extracted from the real-time twin image, while dynamic features, including personnel locations and equipment operating parameters, are output as a standardized risk feature vector.

5. The real-time early warning system for safety risks at construction sites based on digital twins according to claim 4, characterized in that, The digital twin dynamic mapping layer also includes a lightweight rendering module, which receives the real-time updated twin model, performs lightweight processing, and outputs an interactive rendering model through a 3D engine.

6. The real-time early warning system for safety risks at construction sites based on digital twins as described in claim 1, characterized in that, The security risk identification and early warning layer assesses the risk level and generates graded early warning signals based on the assessment results, including the following: The system receives the risk feature vector output from the digital twin dynamic mapping layer, optimizes the anchor box YOLOv8 model through K-means clustering, and identifies risk targets such as personnel violations and equipment malfunctions by combining the attention mechanism, and outputs the entity risk results. Based on the entity risk results, the weights of the indicators are determined by the analytic hierarchy process (AHP) algorithm, the membership degree is determined by fuzzy numbers, and the quantitative risk level is calculated and output by the fuzzy hierarchical comprehensive evaluation model. Receive the quantified risk level, trigger the early warning according to the preset rules, and output the risk-related information simultaneously, including the risk location and risk type.

7. The real-time early warning system for safety risks at construction sites based on digital twins as described in claim 1, characterized in that, The early warning output and response layer's generation of handling plans and completion of closed-loop updates include the following: Receive early warning signals and output them synchronously through the output device; Receive risk type and level information, call the preset handling solution library through the rule engine, and output differentiated handling strategies; The results of the actions, whether manually entered or automatically monitored, are fed back to the digital twin dynamic mapping layer to update the model status and store the action records.

8. The real-time early warning system for safety risks at construction sites based on digital twins according to claim 7, characterized in that, In the warning output and response layer, the rule engine has a built-in preset solution library containing the correspondence between risk types and disposal measures. The solution library is updated according to changes in the construction scenario. Before the disposal strategy is output, it is sent to the digital twin model for simulation verification of feasibility. If it is feasible, it is output; if it is not feasible, it is terminated.

9. The real-time early warning system for safety risks at construction sites based on digital twins as described in claim 1, characterized in that, It also includes an edge and cloud collaboration module, which serves as a support layer for data transmission and storage, enabling layered processing: This includes edge gateways, which are deployed at construction sites to receive multi-source heterogeneous data output from the data acquisition layer, perform local simplification processing, and output the simplified data to the cloud, reducing transmission pressure. This includes cloud nodes, which use elastic servers to receive simplified data and operational logs from the edge gateway, store twin models, handling records, and other data, and output storage services to support interactions between different layers.

10. A computer storage medium, characterized in that: It stores a program that, when executed by a processor, implements the real-time early warning system for safety risks at construction sites based on digital twins, as described in claims 1-9.

Images