Intelligent construction site construction risk early warning system and method based on BIM technology

Abstract

The invention discloses an intelligent construction site construction risk early warning system and method based on a BIM technology, and relates to the technical field of the Internet of Things. Multi-dimensional data of the field environment, structure, equipment and personnel are collected through a multi-source sensor network and are preprocessed; mapping the data to a BIM model, analyzing a spatial interaction relationship among personnel, machinery and environment by using a space-time diagram neural network, and identifying a periodic risk through an autocorrelation algorithm; calculating an index weight in combination with a dynamic weight distribution algorithm, and constructing a risk value calculation model; risk levels are analyzed and judged according to three conditions that personnel are at a mechanical operation position, under mechanical operation and no personnel are on site; a grading response strategy is adopted according to the risk grade; and optimizing an early warning threshold value through a threshold value judgment algorithm and an LSTM model by using a block chain evidence storage risk event, and iteratively updating a prediction algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of Internet of Things, and in particular to a smart construction site construction risk early warning system and method based on BIM technology. Background Art

[0002] With the rapid development of the construction industry, smart construction site technology has become a crucial tool for improving construction safety management. Traditional construction safety management relies primarily on manual inspections and empirical judgment, which can lead to delayed response times and low efficiency. Existing sensor networks often monitor only a single dimension and lack the ability to collaboratively collect and comprehensively analyze multi-source data, making it difficult to comprehensively assess complex risks at construction sites.

[0003] The lack of real-time and accuracy in risk warnings is another prominent issue. Traditional methods based on static threshold judgments cannot dynamically adapt to changes in the construction environment, and their ability to identify risks of interaction between people, machinery, and the environment is limited, resulting in high rates of false alarms or missed reports. In terms of response strategies, existing technologies rely heavily on manual decision-making, making it difficult to achieve a fast and accurate emergency response, especially in scenarios with multiple risk couplings, where the optimal disposal plan cannot be automatically generated. Insufficient use of historical data is also a limiting factor. Traditional database storage methods have problems such as easy data tampering and difficulty in tracing data. In addition, the lack of in-depth mining of historical data makes it difficult to continuously optimize the early warning model. Summary of the Invention

[0004] The purpose of the present invention is to provide a smart construction site construction risk early warning system and method based on BIM technology to solve the problems raised in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solutions: In a first aspect, the present invention provides a smart construction site construction risk early warning method based on BIM technology, comprising: Through a multi-source sensor network, multi-dimensional data from the construction site is collected, including environmental parameters, structural safety indicators, equipment operating status, and personnel behavior data. The raw data is filtered, de-noised, and standardized pre-processed. Map sensor data to the BIM 3D model, use a spatiotemporal graph neural network to analyze the spatial interactions between people, machinery, and the environment, and identify periodic risks through an autocorrelation algorithm. Combined with a dynamic weight allocation algorithm, calculate the weights of each indicator and construct a risk value calculation model. When a person is at the machine operating position, the analysis is conducted. When a single person is operating, the risk level is determined based on the operation time, machine status, and the person's biological characteristics. When multiple people are operating collaboratively, the risk level is determined based on the number of people, operating status, and environmental parameters. When a person is below the machine, the analysis is conducted. When surrounding machines are in operation, the risk level is determined based on the spatial relationship between the density of people and the machine's operating trajectory, as well as the dwell time. When no personnel are present, the analysis is conducted and the risk level is determined based on the degree of deviation between the machine's operating parameters and the preset safety thresholds under normal operating conditions and abnormal environmental conditions. Adopt a graded response strategy based on risk levels. When the risk level is low, the BIM model is used to mark and alert the responsible personnel. When the risk level is medium, partial operations are suspended and on-site sound and light alarms are activated. When the risk level is high, forced shutdown is implemented, emergency broadcasts are activated, and evacuation plans are implemented. The entire process of risk events is recorded through blockchain technology, the warning threshold is optimized based on the threshold judgment algorithm, and the prediction algorithm is iteratively updated using the LSTM model to analyze historical data.

[0006] In conjunction with the first aspect, in a first implementation of the first aspect of the present application, the multi-dimensional data of the construction site is collected through a multi-source sensor network, including environmental parameters, structural safety indicators, equipment operating status, and personnel behavior data, and the raw data is filtered, denoised, and standardized preprocessing is performed, including: Adopt multi-source sensor networking deployment, including environmental monitoring sensors, structural safety monitoring sensors, equipment operation status sensors, and personnel positioning sensors, to synchronously collect multi-dimensional data from the construction site to ensure the temporal and spatial consistency of the data; Environmental monitoring sensors include temperature and humidity sensors, wind speed sensors, PM2.5 sensors, noise sensors, and light intensity sensors; structural safety monitoring sensors include strain sensors, tilt sensors, and vibration sensors; equipment operation status sensors include current and voltage sensors, speed sensors, and oil pressure sensors; personnel positioning sensors include UWB positioning base stations and helmet sensors; The collected data is preprocessed, and the Kalman filter algorithm is used to eliminate the random noise of the environmental parameter data; the wavelet transform method is used to denoise the structural safety monitoring data; the equipment operation status data is smoothed by sliding window mean filtering; and the DBSCAN clustering algorithm is used to clean the abnormal points in the personnel positioning data.

[0007] In combination with the first aspect, in a second implementation of the first aspect of the present application, mapping sensor data to a BIM three-dimensional model, using a spatiotemporal graph neural network to analyze the spatial interaction between personnel, machinery, and the environment, and identifying periodic risks through an autocorrelation algorithm include: A BIM 3D model spatial coordinate system based on the IFC standard is constructed. A coordinate conversion algorithm is used to map the spatiotemporal data collected by sensors to model components. Static component data is associated with unique IDs, and dynamic object data is matched using a spatial interpolation algorithm. A spatiotemporal graph neural network is used to establish an interactive relationship analysis model. With people and machinery as nodes, dynamic edge connections are constructed based on the correlation between distance and motion trajectory. Environmental parameters are introduced as edge feature weights in the graph attention network. Short-term spatiotemporal features are extracted using 3D convolution kernels, and long-term dependencies are captured using a GRU module. The autocorrelation algorithm is used to identify periodic risk patterns, and the fast Fourier transform is used to extract the daily and weekly periodic components of regular operation data. For non-fixed interval events, the sliding window autocorrelation function combined with the dynamic time warping algorithm is used to match historical waveforms, and the abnormal deviation of the periodic pattern is predicted through the LSTM network.

[0008] In conjunction with the first aspect, in a third implementation of the first aspect of the present application, the method of combining a dynamic weight allocation algorithm to calculate the weights of each indicator and construct a risk value calculation model includes: A hybrid weight calculation framework based on the analytic hierarchy process and entropy weight method was established to construct a three-layer indicator system. The target layer is the comprehensive risk value, the criterion layer includes personnel, machinery, and environmental indicators, and the indicator layer is the specific monitoring parameters. Set the weights of each indicator and parameter at the criterion layer and indicator layer, set the corresponding risk value rules for each parameter, calculate the risk value of the criterion layer based on the parameter weights, and then calculate the comprehensive risk value R based on the weights of the criterion layer. The formula is as follows: ; Among them, W p is the personnel risk weight, W m is the mechanical risk weight, W e is the environmental risk weight, R p is the personnel risk value, R m is the mechanical risk value, R e is the environmental risk value; The risk levels are divided into low risk, medium risk and high risk, and the judgment is made based on the comprehensive risk value obtained through the user's preset risk level threshold range.

[0009] In combination with the first aspect, in a fourth implementation of the first aspect of the present application, the analysis is performed when a person is in a machine operating position. When a single person is operating, risk grading is performed based on the operation duration, machine status, and person's biometrics. When multiple people are operating collaboratively, the risk level is determined based on the number of people, operating status, and environmental parameters, including: When a single operator is operating, sensors monitor the operator's operating time and obtain data on the machine's operating status, including vibration, temperature, and speed. Heart rate and fatigue monitoring are used to collect the operator's physiological data and analyze the operator's physical condition and fatigue level. The weighted summation of the risk factors for operating time, machine status, and operator biometrics is used to determine the overall risk value for the single operator. The final risk level is then determined based on pre-set risk classification criteria. When multiple personnel collaborate in an operation, personnel positioning sensors are used to count the number of personnel in the operation area, monitor the operational coordination among multiple personnel, including operation intervals and movement synchronization, determine whether there are operational conflicts or improper coordination, and collect environmental parameters of the operation area, including temperature, humidity, and noise. Combined with the preset environmental risk threshold, it is determined whether the environment is suitable for operation; the risk coefficients of the number of personnel, operating status, and environmental parameters are weighted and summed to obtain a comprehensive risk value for the collaborative operation of multiple personnel, and the risk level is determined according to the preset risk level classification standard.

[0010] In conjunction with the first aspect, in a fifth implementation of the first aspect of the present application, the analysis is performed when personnel are below the machine operation, and the risk level is determined based on the spatial relationship between personnel density and the machine operation trajectory and the residence time when the surrounding machinery is operating, including: Monitor the number and distribution of personnel in the area below mechanical operations, obtain personnel location information through UWB positioning, calculate the personnel density per unit area, obtain the machine's operating trajectory and location information through the positioning device installed on the machine, integrate the machine's location information with the BIM model, analyze the spatial relationship between the mechanical operation trajectory and the area where the personnel are located, determine whether the mechanical operation trajectory intersects or overlaps with the personnel location, and the degree of intersection or overlap, and monitor the time personnel spend in the area below the mechanical operation; combine the analysis results of personnel density and mechanical operation trajectory to comprehensively assess the risk level of personnel when they are below the mechanical operation; The risk assessment results of personnel density, mechanical operation trajectory and residence time are weighted and summed to obtain the comprehensive risk value of personnel under mechanical operation. The risk level is determined according to the preset risk level classification standard.

[0011] In conjunction with the first aspect, in a sixth implementation of the first aspect of the present application, the analysis performed when no personnel are present, and the risk level determined based on the degree of deviation between the machine operating parameters and the preset safety threshold under normal operating conditions and abnormal environmental conditions, includes: Monitor the operating parameters of the machinery when it is unmanned, collect parameter data through the equipment operation status sensor, and analyze the data; preset safety thresholds for the machinery operation parameters, with safety thresholds divided into different standards for normal operating environments and abnormal environments, compare the monitored machinery operation parameters with the preset safety thresholds, calculate the degree of deviation between the parameter data and the safety thresholds, and divide the machinery operation status into different risk levels.

[0012] In conjunction with the first aspect, in a seventh implementation of the first aspect of the present application, a hierarchical response strategy is adopted according to the risk level. When the risk level is low, the responsible personnel are reminded through BIM model annotation; when the risk level is medium, the operation is partially suspended and the on-site sound and light alarm is activated; when the risk level is high, the equipment is forced to stop, the emergency broadcast is activated, and the evacuation plan is implemented, including: When the risk level is determined to be low, the corresponding area in the BIM model is highlighted, a warning icon is displayed in the model view, and an early warning message is generated. The responsible personnel are notified via SMS, APP push, or on-site display screens to remind them of the potential risks in the current area and to take appropriate measures. When the risk level is determined to be medium, a local suspension mechanism is triggered. The affected area is delineated using the Delaunay triangulation algorithm, operations in the relevant area or equipment are suspended, and the sound and light alarms are activated. Responsible personnel assess the risk situation and take necessary measures. When the risk level is determined to be high, the forced shutdown mechanism is triggered, the operation of related equipment is stopped, the power is cut off, and emergency evacuation instructions are issued through emergency broadcasting. The optimal evacuation path is generated based on the A* algorithm, personnel are guided to evacuate, and the emergency plan is activated.

[0013] In conjunction with the first aspect, in an eighth implementation of the first aspect of this application, the whole-process risk event is recorded through blockchain technology, the warning threshold is optimized based on the threshold judgment algorithm, and the prediction algorithm is iteratively updated by analyzing historical data using the LSTM model, including: The key data in the risk warning process is packaged into data blocks, including the time, location, type, level, personnel involved, equipment information, response measures and results of the risk event. The data blocks are added to the blockchain in chronological order, and the recording and verification of risk events are automatically executed through smart contracts. Collect historical risk event data, including pre-risk monitoring data, risk level determination results, and actual risk situations; analyze the relationship between different risk indicators in historical data and actual risk levels; use threshold judgment algorithms to evaluate the performance of current warning thresholds and adjust the warning thresholds for each risk indicator; Use the LSTM network to establish a spatiotemporal series prediction model, analyze historical risk event data, use new risk event data as training samples, input the LSTM model for training, and use cross-validation and regularization techniques to prevent model overfitting. The trained LSTM model is applied to risk warning, and the risk level is predicted based on the current monitoring data, providing a basis for taking preventive measures in advance.

[0014] In a second aspect, the present invention provides a smart construction site risk warning system based on BIM technology, comprising: Data acquisition and perception module: This includes a multi-source sensor networking unit and a data pre-processing unit. The multi-source sensor networking unit uses multiple sensors to synchronously collect multi-dimensional data from the construction site, ensuring the temporal and spatial consistency of the data. The data pre-processing unit performs filtering, noise reduction, and standardization on the collected raw data. Data Mapping and Analysis Module: This module includes a BIM model mapping unit, a spatiotemporal graph neural network analysis unit, and an autocorrelation analysis unit. The BIM model mapping unit maps pre-processed sensor data to the BIM 3D model to associate the data with the model. The spatiotemporal graph neural network analysis unit uses a spatiotemporal graph neural network to analyze the spatial interactions between people, machinery, and the environment, identifying potential safety risks and abnormal behaviors. The autocorrelation analysis unit performs autocorrelation analysis on historical data to identify periodic risk patterns. Risk value calculation and assessment module: includes a dynamic weight allocation unit, a risk value calculation unit, and a risk level determination unit. The dynamic weight allocation unit combines the analytic hierarchy process (AHP) and the entropy weight method to construct a three-tier indicator system, calculate indicator weights and the comprehensive risk value R. The risk value calculation unit presets risk level classification standards and divides the comprehensive risk value R into low risk, medium risk, and high risk levels. The risk level determination unit determines the final risk level based on the classification standards and the calculated comprehensive risk value. Warning information release and response module: includes a warning information release unit and an emergency plan triggering unit. The warning information release unit releases warning information based on risk levels through BIM model annotation, sound and light alarms, emergency broadcasts, and information push. The emergency plan triggering unit triggers the corresponding emergency plan based on risk levels. Data storage analysis and optimization module: includes blockchain evidence unit, threshold optimization unit and LSTM model training and application unit; among them, the blockchain evidence unit packages the data in the risk warning process into data blocks, adds them to the blockchain in chronological order, and performs recording and verification of risk events; the threshold optimization unit analyzes the relationship between different risk indicators in historical data and the actual risk level, evaluates the performance of the current warning threshold through algorithms, and adjusts the warning threshold of each risk indicator; the LSTM model training and application unit uses the LSTM network to establish a spatiotemporal sequence prediction model, analyzes historical risk event data, trains the LSTM model, and applies the trained LSTM model to real-time risk warning.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention maps sensor data to a BIM three-dimensional model, uses a spatiotemporal graph neural network to analyze the spatial interaction between personnel, machinery, and the environment, and identifies periodic risks through an autocorrelation algorithm.

[0016] 2. The present invention establishes a hybrid weight calculation framework based on the hierarchical analysis method and the entropy weight method, constructs a three-layer indicator system, sets the weights of each indicator and parameter, sets corresponding risk value rules for each parameter, calculates the comprehensive risk value R, and performs risk level classification.

[0017] 3. The present invention analyzes the situations when personnel are at the machine operating position, personnel are under the machine operation, and when there are no personnel present, compares the relevant parameter data with the preset threshold value, and obtains the comprehensive risk value by weighted summation according to the risk coefficient to determine the risk level. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Fig. 1 This is a schematic diagram of the steps of the smart construction site construction risk early warning method based on BIM technology of the present invention; Fig. 2 This is a system structure diagram of the smart construction site construction risk early warning system based on BIM technology of the present invention. DETAILED DESCRIPTION

[0019] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0020] Example: Figs. 1-2 As shown, the present invention provides a technical solution. like Fig. 1As shown in the schematic diagram of the steps of the smart construction site construction risk early warning method based on BIM technology, the present invention provides a smart construction site construction risk early warning method based on BIM technology, comprising: Step S100: Collect multi-dimensional data of the construction site through a multi-source sensor network, including environmental parameters, structural safety indicators, equipment operating status, and personnel behavior data, and perform filtering, noise reduction, and standardization preprocessing on the raw data; Specifically, a multi-source sensor network is deployed, including environmental monitoring sensors, structural safety monitoring sensors, equipment operation status sensors, and personnel positioning sensors, to synchronously collect multi-dimensional data from the construction site to ensure the temporal and spatial consistency of the data; Environmental monitoring sensors include temperature and humidity sensors, wind speed sensors, PM2.5 sensors, noise sensors, and light intensity sensors; structural safety monitoring sensors include strain sensors, tilt sensors, and vibration sensors; equipment operation status sensors include current and voltage sensors, speed sensors, and oil pressure sensors; personnel positioning sensors include UWB positioning base stations and helmet sensors; The collected data is preprocessed, and the Kalman filter algorithm is used to eliminate the random noise of the environmental parameter data; the wavelet transform method is used to denoise the structural safety monitoring data; the equipment operation status data is smoothed by sliding window mean filtering; and the DBSCAN clustering algorithm is used to clean the abnormal points in the personnel positioning data.

[0021] In a specific embodiment, a construction site of a large commercial complex is selected for data collection. Various sensors, including environmental monitoring sensors, structural safety monitoring sensors, equipment operation status sensors, and personnel positioning sensors, are deployed in key areas of the construction site, such as the edge of a deep foundation pit, aerial work platforms, material storage areas, and crowded passages, to collect data from the construction site.

[0022] The data collected at 3:30 PM on March 14, 2025, was selected. The environmental monitoring sensor collected the following data: temperature 32°C, humidity 68%, wind speed 5 m / s, PM2.5 concentration 35 µg / m³, noise 75 dB, and light intensity 500 lux. The data feedback from the structural safety monitoring sensor showed a strain value of 0.0005, a tilt of 2°, a vibration frequency of 10 Hz, and an amplitude of 0.1 m / s. The equipment operation status sensor collected current 10 A, voltage 220 V, speed 1500 RPM, and oil pressure 5 bar. The personnel positioning sensor captured a worker with tag ID 001, whose coordinates were [10 m, 5 m] and whose helmet was worn normally.

[0023] The above data were preprocessed by using the Kalman filter algorithm to filter the temperature data to 31.8°C and the humidity to 67.5%. The wavelet transform method was used to reduce the strain value to 0.0004 and the amplitude to 0.05m / s. The current was averaged to 9.8A, the voltage to 218V, and the speed to 1480RPM through sliding window mean filtering. The DBSCAN clustering algorithm was used to correct the personnel coordinates to [10.2m, 4.8m].

[0024] Step S200: Mapping sensor data to a BIM 3D model, using a spatiotemporal graph neural network to analyze the spatial interactions between people, machinery, and the environment, and identifying periodic risks through an autocorrelation algorithm. Combining a dynamic weight allocation algorithm, the weights of each indicator are calculated to construct a risk value calculation model. Specifically, a BIM three-dimensional model spatial coordinate system based on the IFC standard is constructed. A coordinate conversion algorithm is used to map the spatiotemporal data collected by sensors to model components. Static component data is associated with unique IDs, and dynamic object data is matched using a spatial interpolation algorithm. A spatiotemporal graph neural network is used to establish an interactive relationship analysis model. With people and machinery as nodes, dynamic edge connections are constructed based on the correlation between distance and motion trajectory. Environmental parameters are introduced as edge feature weights in the graph attention network. Short-term spatiotemporal features are extracted using 3D convolution kernels, and long-term dependencies are captured by combining the GRU module. The autocorrelation algorithm is used to identify periodic risk patterns, and the fast Fourier transform is used to extract the daily and weekly periodic components of regular operation data. For non-fixed interval events, the sliding window autocorrelation function combined with the dynamic time warping algorithm is used to match historical waveforms, and the abnormal deviation of the periodic pattern is predicted through the LSTM network.

[0025] Furthermore, a hybrid weight calculation framework based on the analytic hierarchy process and entropy weight method is established to construct a three-layer indicator system, where the target layer is the comprehensive risk value, the criterion layer includes personnel, machinery, and environmental indicators, and the indicator layer is the specific monitoring parameters; Set the weights of each indicator and parameter at the criterion layer and indicator layer, set the corresponding risk value rules for each parameter, calculate the risk value of the criterion layer based on the parameter weights, and then calculate the comprehensive risk value R based on the weights of the criterion layer. The formula is as follows: ; Among them, W p is the personnel risk weight, W m is the mechanical risk weight, W e is the environmental risk weight, R p is the personnel risk value, R m is the mechanical risk value, R e is the environmental risk value; The risk levels are divided into low risk, medium risk and high risk, and the judgment is made based on the comprehensive risk value obtained through the user's preset risk level threshold range.

[0026] In a specific embodiment, a high-rise building construction site was selected for testing, a complete sensor network was deployed, and a BIM 3D model based on the IFC standard was established. Data collected at 11:45:29 on March 24, 2025 was selected: the environmental monitoring sensor showed a temperature of 28°C, a humidity of 60%, a wind speed of 3m / s, a PM2.5 concentration of 25μg / m³, a noise level of 70dB, and a light intensity of 400lux; the structural safety monitoring sensor reported a strain value of 0.0003, an inclination of 1.5°, a vibration frequency of 8Hz, and an amplitude of 0.05m / s; the equipment operation status sensor showed a tower crane current of 15A, a voltage of 380V, a speed of 1200RPM, and an oil pressure of 4bar; and the personnel positioning sensor captured the real-time coordinates and helmet wearing status of 10 workers. Worker A's coordinates were [15m, 20m, 10m], and his helmet was worn normally.

[0027] Through the coordinate conversion algorithm, the data is accurately mapped to the BIM three-dimensional model. Static components including beams and columns are associated with sensor data through unique IDs, and dynamic objects including tower cranes and workers are matched to the corresponding positions in the model through the spatial interpolation algorithm. A spatiotemporal graph neural network analysis model is adopted, with workers and tower cranes as nodes. Dynamic edge connections are constructed based on the correlation between real-time distance and motion trajectory. Environmental parameters including temperature and humidity are used as edge feature weights of the graph attention network. Short-term spatiotemporal features are extracted through 3D convolution kernels, and long-term dependencies are captured by combining with the GRU module to analyze the spatial interaction relationship between people, machinery and the environment.

[0028] The autocorrelation algorithm is used to identify periodic risks in the collected data. The fast Fourier transform is used to extract the daily and weekly components of the regular operation data of the tower crane. The sliding window autocorrelation function combined with the dynamic time warping algorithm is used to match the historical waveform of the workers' non-fixed interval movement events. The LSTM network is used to predict abnormal deviations in the periodic pattern to identify potential safety risks.

[0029] Furthermore, a hybrid weight calculation framework based on the analytic hierarchy process and entropy weight method is established to construct a three-layer indicator system: the target layer is the comprehensive risk value, the criterion layer includes personnel risk, mechanical risk and environmental risk, and the indicator layer is the specific monitoring parameters; the weights of each indicator and parameter in the criterion layer and the indicator layer are set, and the personnel risk weight W p is 0.4, mechanical risk weight W m The environmental risk weight is 0.3, e0.3, set the score standard of each parameter, and calculate the risk value of the criterion layer according to the parameter, and then calculate the comprehensive risk value R according to the weight of the criterion layer, and obtain the personnel risk value R p 0.6, the mechanical risk value R m 0.5, the environmental risk value R e 0.4, the comprehensive risk value R is calculated by the following formula: ; R=0.51 is calculated, according to the risk level threshold range preset by the user, 0-0.3 is judged as low risk, 0.3-0.7 is judged as medium risk, and 0.7-1 is judged as high risk, and the calculated comprehensive risk value 0.51 belongs to the medium risk range; Step S300: When the personnel are in the mechanical operation position, the analysis is carried out, the single personnel operation is combined with the operation time, the mechanical state and the personnel biological characteristics to carry out risk grading, the number of personnel, the operation state and the environmental parameters are combined to determine the risk level when multiple personnel operate cooperatively; When the personnel are under the mechanical operation, the analysis is carried out, and the risk level is determined according to the spatial relationship between the personnel density and the mechanical operation track and the staying time when the surrounding machinery runs; When there is no personnel present, the analysis is carried out, and the risk level is determined according to the deviation degree of the mechanical running parameters from the preset safety threshold in the normal operation environment and the abnormal environmental conditions; Specifically, when single personnel operates, the operation time of the personnel is monitored through the sensor, the running state data of the machinery is obtained, including vibration, temperature and speed, the physiological data of the personnel is collected through heart rate monitoring and fatigue monitoring, and the body condition and fatigue degree of the personnel are analyzed; The risk coefficients of the operation time, the mechanical state and the personnel biological characteristics are weighted and summed to obtain the comprehensive risk value of the single personnel operation, and the final risk level is determined according to the preset risk level division standard; When multiple personnel operate cooperatively, the number of personnel in the operation area is counted through the personnel positioning sensor, the operation coordination between multiple personnel is monitored, including operation interval and action synchronism, whether there is operation conflict or improper cooperation is judged, the environmental parameters of the operation area are collected, including temperature, humidity and noise, whether the environment is suitable for operation is judged in combination with the preset environmental risk threshold; The risk coefficients of the number of personnel, the operation state and the environmental parameters are weighted and summed to obtain the comprehensive risk value of the multiple personnel cooperative operation, and the risk level is determined according to the preset risk level division standard.

[0030] Furthermore, the number and distribution of personnel in the area below the machine operation are monitored. Personnel location information is obtained through UWB positioning, and the personnel density per unit area is calculated. The machine's operation trajectory and location information are obtained through the positioning device installed on the machine. The machine's location information is integrated with the BIM model to analyze the spatial relationship between the machine's operation trajectory and the area where the personnel are located. It is determined whether the machine's operation trajectory intersects or overlaps with the personnel's location, and the degree of intersection or overlap. The length of time personnel stay in the area below the machine operation is monitored. Combined with the analysis results of personnel density and machine operation trajectory, a comprehensive assessment of the risk level of personnel when they are below the machine operation is made. The risk assessment results of personnel density, mechanical operation trajectory and residence time are weighted and summed to obtain the comprehensive risk value of personnel under mechanical operation. The risk level is determined according to the preset risk level classification standard.

[0031] Furthermore, the operating parameters of the machinery are monitored when there is no one operating it, and parameter data is collected through the equipment operating status sensor, and the data is analyzed; the safety thresholds of the machinery operating parameters are preset, and the safety thresholds are divided into different standards for normal operating environments and abnormal environments. The monitored machinery operating parameters are compared with the preset safety thresholds, and the degree of deviation between the parameter data and the safety thresholds is calculated, and the machinery operating status is divided into different risk levels.

[0032] In a specific embodiment, a construction site in Shanghai was selected to conduct risk assessments for different scenarios. In a single-person operation scenario, the sensor monitored the operator's operation time for 4 hours, and the machine's operating status data showed a vibration value of 3.2m / s², a temperature of 75°C, and a speed of 1500rpm. At the same time, the operator's physiological data was collected through heart rate monitoring and fatigue monitoring, with a heart rate of 85 beats / minute and a fatigue index of 0.7. The risk coefficients of the operation time, machine status, and personnel's biological characteristics were weighted and summed, and the comprehensive risk value of the single-person operation was obtained by the formula to be 0.65. According to the preset risk level classification standard, the risk level was determined to be medium risk. In a multi-person collaborative operation scenario, the personnel positioning sensor counted five people in the operation area. Monitoring showed good coordination among the multiple personnel, with operation intervals and movement synchronization within normal ranges. The collected environmental parameters for the operation area showed a temperature of 28°C, a humidity of 60%, and a noise level of 65dB. By weighting the risk coefficients for the number of personnel, operating status, and environmental parameters, the formula yielded a comprehensive risk value of 0.45 for multi-person collaborative operation. Based on the preset risk level classification criteria, the risk level was determined to be low. In the area below the mechanical operation, personnel location information is obtained through UWB positioning, and the personnel density per unit area is calculated to be 0.8 people / m². The operation trajectory and position information of the machinery are obtained through the positioning device installed on the machinery. After integration with the BIM model, analysis shows that there is partial intersection between the mechanical operation trajectory and the area where the personnel are located, with an intersection degree of 30%. The monitored stay time of personnel in the area below the mechanical operation is 20 minutes. Combined with the analysis results of personnel density and mechanical operation trajectory, the comprehensive risk value of personnel under the mechanical operation is comprehensively assessed to be 0.55. According to the preset risk level classification standard, the risk level is determined to be medium risk.

[0033] When no one is operating, the machine's operating parameters are collected through the equipment operation status sensor, with a vibration value of 2.8m / s², a temperature of 70°C, and a speed of 1400rpm. The preset safety thresholds for the machine's operating parameters under normal operating conditions are vibration values ​​≤3.0m / s², temperature ≤80°C, and speed ≤1600rpm; under abnormal conditions, they are vibration values ​​≤3.5m / s², temperature ≤90°C, and speed ≤1800rpm. The monitored machine operating parameters are compared with the preset safety thresholds, and the calculated deviation between the parameter data and the safety threshold is 0.1, classifying the machine's operating status as a low-risk level.

[0034] Step S400: Adopt a graded response strategy based on the risk level. When the risk level is low, the BIM model is marked to alert the responsible personnel; when the risk level is medium, the operation is partially suspended and the on-site sound and light alarm is activated; when the risk level is high, the equipment is forced to stop, the emergency broadcast is activated, and the evacuation plan is implemented; Specifically, when the risk level is determined to be low, the corresponding area in the BIM model will be highlighted, a warning icon will be displayed in the model view, and an early warning message will be generated. The responsible personnel will be notified via SMS, APP push or on-site display screens to remind them to pay attention to the potential risks in the current area and take corresponding measures; When the risk level is determined to be medium, a local suspension mechanism is triggered. The affected area is delineated using the Delaunay triangulation algorithm, operations in the relevant area or equipment are suspended, and the sound and light alarms are activated. Responsible personnel assess the risk situation and take necessary measures. When the risk level is determined to be high, the forced shutdown mechanism is triggered, the operation of related equipment is stopped, the power is cut off, and emergency evacuation instructions are issued through emergency broadcasting. The optimal evacuation path is generated based on the A* algorithm, personnel are guided to evacuate, and the emergency plan is activated.

[0035] In a specific embodiment, a corresponding graded response strategy is adopted according to actual monitoring data and risk assessment results; When a single person is operating, the monitoring data shows that the operation time is 2 hours, the mechanical vibration value is 2.5m / s, the ambient temperature is 25°C, and the person's heart rate is 75 beats / minute. The calculated comprehensive risk value is 0.3, which is judged to be low risk. In the BIM model, the operation area is highlighted in orange, and the "Attention" warning icon is displayed in the model view, generating an early warning message: "There is a potential risk in the operation area, please strengthen monitoring!" This information is notified to the responsible person Zhang Wei via SMS, APP push and on-site display screen, reminding him to pay attention to the current situation in the area and recommending an increase in the frequency of inspections.

[0036] When multiple personnel are working together, monitoring data shows that there are 5 personnel, the mechanical vibration value suddenly rises to 3.8m / s, and the ambient humidity reaches 85%. The comprehensive risk value is calculated to be 0.6, which is judged to be medium risk, triggering a local suspension of operations. The Delaunay triangulation algorithm is used to define the affected area with a radius of 5 meters, and all operations in the area are suspended. At the same time, the sound and light alarm device is activated, emitting yellow flashes and continuous alarm sounds. The responsible person Xiao Li immediately rushed to the scene, assessed the risk situation, and decided to perform temporary maintenance on the machinery and adjust the operation plan to reduce the density of personnel.

[0037] When no one was operating, monitoring data showed that the machine temperature suddenly rose to 95°C, exceeding the safety threshold of 90°C under abnormal conditions. The calculated comprehensive risk value was 0.9, which was judged to be high risk. The forced shutdown mechanism was immediately triggered, the operation of related equipment was stopped, and the power was cut off. The emergency broadcast system was activated, and an emergency evacuation order was issued: "Emergency, all personnel, please evacuate immediately!" Based on the A* algorithm, the optimal evacuation path from each work point to the safe exit was generated and displayed on the BIM model and on-site display screen. At the same time, the emergency plan was activated, and the safety management personnel organized the orderly evacuation of personnel, which was completed within 5 minutes. Subsequently, a professional team conducted a comprehensive inspection and maintenance of the equipment.

[0038] Step S500: Use blockchain technology to store risk events throughout the entire process, optimize the warning threshold based on the threshold judgment algorithm, and use the LSTM model to analyze historical data and iteratively update the prediction algorithm.

[0039] Specifically, key data in the risk warning process is packaged into data blocks, including the time, location, type, level, personnel involved, equipment information, response measures and results of the risk event. The data blocks are added to the blockchain in chronological order, and the recording and verification of risk events are automatically executed through smart contracts. Collect historical risk event data, including pre-risk monitoring data, risk level determination results, and actual risk situations; analyze the relationship between different risk indicators in historical data and actual risk levels; use threshold judgment algorithms to evaluate the performance of current warning thresholds and adjust the warning thresholds for each risk indicator; Use the LSTM network to establish a spatiotemporal series prediction model, analyze historical risk event data, use new risk event data as training samples, input the LSTM model for training, and use cross-validation and regularization techniques to prevent model overfitting. The trained LSTM model is applied to risk warning, and the risk level is predicted based on the current monitoring data, providing a basis for taking preventive measures in advance.

[0040] In a specific embodiment, a mechanical failure risk event occurred in area A of the construction site. The type was equipment overheating, and the level was determined to be medium risk. The operator involved was Li Si, and the equipment number was M20230515. The response measure was to immediately shut down and cool the equipment. The result was that the accident was successfully avoided. The data was added to the blockchain in chronological order, and the records and verification were automatically executed through smart contracts to ensure the data's immutability and traceability.

[0041] Historical risk event data was collected. A total of 100 risk events occurred in the past year, of which 60 were related to excessive temperature, 30 were related to excessive humidity, and 10 were related to excessive concentrations of harmful gases. The relationship between different risk indicators in the data, including temperature, humidity, and harmful gas concentration, and the actual risk level was analyzed. It was found that when the current temperature warning threshold was set to 80°C, the missed alarm rate was 15% and the false alarm rate was 10%. The threshold judgment algorithm was used to evaluate and adjust the warning threshold, and the temperature warning threshold was optimized to 75°C, reducing the missed alarm rate and false alarm rate.

[0042] A spatiotemporal series prediction model was established using the LSTM network. Historical risk event data, including monitoring data before the risk occurred and risk level determination results, was input as training samples. The model was trained using cross-validation and regularization techniques to prevent overfitting. After training, the LSTM model was applied to real-time risk warnings. The risk level was predicted based on the current temperature of 72°C, humidity of 60%, and harmful gas concentration of 0.02%. The model prediction results showed that there was an 80% probability that the temperature would exceed the warning threshold of 75°C in the next 24 hours. It was recommended to take preventive measures in advance, including adding ventilation equipment and adjusting the work plan.

[0043] like Fig. 2 As shown in the system structure diagram of the smart construction site construction risk early warning system based on BIM technology, the present invention provides a smart construction site construction risk early warning system based on BIM technology, including: Data acquisition and perception module: This includes a multi-source sensor networking unit and a data pre-processing unit. The multi-source sensor networking unit uses multiple sensors to synchronously collect multi-dimensional data from the construction site, ensuring the temporal and spatial consistency of the data. The data pre-processing unit performs filtering, noise reduction, and standardization on the collected raw data. Data Mapping and Analysis Module: This module includes a BIM model mapping unit, a spatiotemporal graph neural network analysis unit, and an autocorrelation analysis unit. The BIM model mapping unit maps pre-processed sensor data to the BIM 3D model to associate the data with the model. The spatiotemporal graph neural network analysis unit uses a spatiotemporal graph neural network to analyze the spatial interactions between people, machinery, and the environment, identifying potential safety risks and abnormal behaviors. The autocorrelation analysis unit performs autocorrelation analysis on historical data to identify periodic risk patterns. Risk value calculation and assessment module: includes a dynamic weight allocation unit, a risk value calculation unit, and a risk level determination unit. The dynamic weight allocation unit combines the analytic hierarchy process (AHP) and the entropy weight method to construct a three-tier indicator system, calculate indicator weights and the comprehensive risk value R. The risk value calculation unit presets risk level classification standards and divides the comprehensive risk value R into low risk, medium risk, and high risk levels. The risk level determination unit determines the final risk level based on the classification standards and the calculated comprehensive risk value. Warning information release and response module: includes a warning information release unit and an emergency plan triggering unit. The warning information release unit releases warning information based on risk levels through BIM model annotation, sound and light alarms, emergency broadcasts, and information push. The emergency plan triggering unit triggers the corresponding emergency plan based on risk levels. Data storage analysis and optimization module: includes blockchain evidence unit, threshold optimization unit and LSTM model training and application unit; among them, the blockchain evidence unit packages the data in the risk warning process into data blocks, adds them to the blockchain in chronological order, and performs recording and verification of risk events; the threshold optimization unit analyzes the relationship between different risk indicators in historical data and the actual risk level, evaluates the performance of the current warning threshold through algorithms, and adjusts the warning threshold of each risk indicator; the LSTM model training and application unit uses the LSTM network to establish a spatiotemporal sequence prediction model, analyzes historical risk event data, trains the LSTM model, and applies the trained LSTM model to real-time risk warning.

[0044] 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 embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims, not the foregoing description, and all variations within the meaning and range of equivalents of the claims are intended to be included therein. Any reference sign in a claim should not be construed as limiting the claim to which it relates.

Claims

1. The smart construction site construction risk warning method based on BIM technology is characterized by: include: Through a multi-source sensor network, multi-dimensional data from the construction site is collected, including environmental parameters, structural safety indicators, equipment operating status, and personnel behavior data. The raw data is filtered, de-noised, and standardized pre-processed. Map sensor data to the BIM 3D model, use a spatiotemporal graph neural network to analyze the spatial interactions between people, machinery, and the environment, and identify periodic risks through an autocorrelation algorithm. Combined with a dynamic weight allocation algorithm, calculate the weights of each indicator and construct a risk value calculation model. When personnel are operating the machine, risk is graded based on the operating time, machine status, and personnel biometrics when the operation is performed by a single operator. When multiple operators are operating collaboratively, the risk level is determined based on the number of personnel, operating status, and environmental parameters. When personnel are under the machine operation, the analysis is carried out. When the surrounding machinery is running, the risk level is determined based on the spatial relationship between personnel density and the machine operation trajectory and the residence time; When no personnel are present, the analysis is conducted and the risk level is determined based on the degree of deviation between the machine operating parameters and the preset safety threshold under normal working conditions and abnormal environmental conditions; Adopt a graded response strategy based on risk levels. When the risk level is low, the BIM model is used to mark and alert the responsible personnel. When the risk level is medium, partial operations are suspended and on-site sound and light alarms are activated. When the risk level is high, forced shutdown is implemented, emergency broadcasts are activated, and evacuation plans are implemented. The entire process of risk events is recorded through blockchain technology, the warning threshold is optimized based on the threshold judgment algorithm, and the prediction algorithm is iteratively updated using the LSTM model to analyze historical data.

2. The BIM-based smart construction site risk early warning method according to claim 1 is characterized in that: The multi-dimensional data of the construction site is collected through a multi-source sensor network, including environmental parameters, structural safety indicators, equipment operating status and personnel behavior data, and the raw data is filtered, denoised and standardized pre-processing is performed, including: Adopt multi-source sensor networking deployment, including environmental monitoring sensors, structural safety monitoring sensors, equipment operation status sensors, and personnel positioning sensors, to synchronously collect multi-dimensional data from the construction site to ensure the temporal and spatial consistency of the data; Environmental monitoring sensors include temperature and humidity sensors, wind speed sensors, PM2.5 sensors, noise sensors, and light intensity sensors; structural safety monitoring sensors include strain sensors, tilt sensors, and vibration sensors; equipment operation status sensors include current and voltage sensors, speed sensors, and oil pressure sensors; personnel positioning sensors include UWB positioning base stations and helmet sensors; The collected data is preprocessed, and the Kalman filter algorithm is used to eliminate the random noise of the environmental parameter data; the wavelet transform method is used to denoise the structural safety monitoring data; the equipment operation status data is smoothed by sliding window mean filtering; and the DBSCAN clustering algorithm is used to clean the abnormal points in the personnel positioning data.

3. The smart construction site construction risk early warning method based on BIM technology according to claim 1 is characterized in that: The sensor data is mapped to the BIM 3D model, and the spatial interaction relationship between personnel, machinery and environment is analyzed using a spatiotemporal graph neural network. The autocorrelation algorithm is used to identify periodic risks, including: A BIM 3D model spatial coordinate system based on the IFC standard is constructed. A coordinate conversion algorithm is used to map the spatiotemporal data collected by sensors to model components. Static component data is associated with unique IDs, and dynamic object data is matched using a spatial interpolation algorithm. A spatiotemporal graph neural network is used to establish an interactive relationship analysis model. With people and machinery as nodes, dynamic edge connections are constructed based on the correlation between distance and motion trajectory. Environmental parameters are introduced as edge feature weights in the graph attention network. Short-term spatiotemporal features are extracted using 3D convolution kernels, and long-term dependencies are captured using a GRU module. The autocorrelation algorithm is used to identify periodic risk patterns, and the fast Fourier transform is used to extract the daily and weekly periodic components of regular operation data. For non-fixed interval events, the sliding window autocorrelation function combined with the dynamic time warping algorithm is used to match historical waveforms, and the abnormal deviation of the periodic pattern is predicted through the LSTM network.

4. The BIM-based smart construction site risk early warning method according to claim 1 is characterized in that: The method combines the dynamic weight allocation algorithm to calculate the weights of each indicator and construct a risk value calculation model, including: A hybrid weight calculation framework based on the analytic hierarchy process and entropy weight method was established to construct a three-layer indicator system. The target layer is the comprehensive risk value, the criterion layer includes personnel, machinery, and environmental indicators, and the indicator layer is the specific monitoring parameters. Set the weights of each indicator and parameter at the criterion layer and indicator layer, set the corresponding risk value rules for each parameter, calculate the risk value of the criterion layer based on the parameter weights, and then calculate the comprehensive risk value R based on the weights of the criterion layer. The formula is as follows: ； Among them, W p is the personnel risk weight, W m is the mechanical risk weight, W e is the environmental risk weight, R p is the personnel risk value, R m is the mechanical risk value, R e is the environmental risk value; The risk levels are divided into low risk, medium risk and high risk, and the judgment is made based on the comprehensive risk value obtained through the user's preset risk level threshold range.

5. The smart construction site construction risk early warning method based on BIM technology according to claim 1 is characterized in that: The analysis is conducted when personnel are at the machine operating position. When a single person is operating, the risk level is graded based on the operation time, machine status, and personnel biometrics. When multiple personnel are operating collaboratively, the risk level is determined based on the number of personnel, operating status, and environmental parameters, including: When a single operator is operating, sensors monitor the operator's operating time and obtain data on the machine's operating status, including vibration, temperature, and speed. Heart rate and fatigue monitoring are used to collect the operator's physiological data and analyze the operator's physical condition and fatigue level. The weighted summation of the risk factors for operating time, machine status, and operator biometrics is used to determine the overall risk value for the single operator. The final risk level is then determined based on pre-set risk classification criteria. When multiple personnel collaborate in an operation, personnel positioning sensors are used to count the number of personnel in the operation area, monitor the operational coordination among multiple personnel, including operation intervals and movement synchronization, determine whether there are operational conflicts or improper coordination, and collect environmental parameters of the operation area, including temperature, humidity, and noise. Combined with the preset environmental risk threshold, it is determined whether the environment is suitable for operation; the risk coefficients of the number of personnel, operating status, and environmental parameters are weighted and summed to obtain a comprehensive risk value for the collaborative operation of multiple personnel, and the risk level is determined according to the preset risk level classification standard.

6. The BIM-based smart construction site risk early warning method according to claim 1 is characterized in that: The analysis is performed when personnel are under the machine operation. When the surrounding machinery is in operation, the risk level is determined based on the spatial relationship between personnel density and the machine operation trajectory and the residence time, including: Monitor the number and distribution of personnel in the area below mechanical operations, obtain personnel location information through UWB positioning, calculate the personnel density per unit area, obtain the machine's operating trajectory and location information through the positioning device installed on the machine, integrate the machine's location information with the BIM model, analyze the spatial relationship between the mechanical operation trajectory and the area where the personnel are located, determine whether the mechanical operation trajectory intersects or overlaps with the personnel location, and the degree of intersection or overlap, and monitor the time personnel spend in the area below the mechanical operation; combine the analysis results of personnel density and mechanical operation trajectory to comprehensively assess the risk level of personnel when they are below the mechanical operation; The risk assessment results of personnel density, mechanical operation trajectory and residence time are weighted and summed to obtain the comprehensive risk value of personnel under mechanical operation. The risk level is determined according to the preset risk level classification standard.

7. The BIM-based smart construction site risk early warning method according to claim 1 is characterized in that: The analysis is conducted when no personnel are present, and the risk level is determined based on the degree of deviation between the machine operating parameters and the preset safety threshold under normal operating conditions and abnormal environmental conditions, including: Monitor the operating parameters of the machinery when it is unmanned, collect parameter data through the equipment operation status sensor, and analyze the data; preset safety thresholds for the machinery operation parameters, with safety thresholds divided into different standards for normal operating environments and abnormal environments, compare the monitored machinery operation parameters with the preset safety thresholds, calculate the degree of deviation between the parameter data and the safety thresholds, and divide the machinery operation status into different risk levels.

8. The BIM-based smart construction site risk early warning method according to claim 1 is characterized in that: The aforementioned graded response strategy is adopted according to the risk level. When the risk level is low, the responsible personnel are reminded through BIM model annotation; when the risk level is medium, the operation is partially suspended and the on-site sound and light alarm is activated; when the risk level is high, the equipment is forced to stop, the emergency broadcast is activated, and the evacuation plan is implemented, including: When the risk level is determined to be low, the corresponding area in the BIM model is highlighted, a warning icon is displayed in the model view, and an early warning message is generated. The responsible personnel are notified via SMS, APP push, or on-site display screens to remind them of the potential risks in the current area and to take appropriate measures. When the risk level is determined to be medium, a local suspension mechanism is triggered. The affected area is delineated using the Delaunay triangulation algorithm, operations in the relevant area or equipment are suspended, and the sound and light alarms are activated. Responsible personnel assess the risk situation and take necessary measures. When the risk level is determined to be high, the forced shutdown mechanism is triggered, the operation of related equipment is stopped, the power is cut off, and emergency evacuation instructions are issued through emergency broadcasting. The optimal evacuation path is generated based on the A* algorithm, personnel are guided to evacuate, and the emergency plan is activated.

9. The BIM-based smart construction site risk early warning method according to claim 1 is characterized in that: The method of using blockchain technology to store risk events throughout the entire process, optimizing warning thresholds based on threshold judgment algorithms, and iteratively updating prediction algorithms using LSTM models to analyze historical data includes: The key data in the risk warning process is packaged into data blocks, including the time, location, type, level, personnel involved, equipment information, response measures and results of the risk event. The data blocks are added to the blockchain in chronological order, and the recording and verification of risk events are automatically executed through smart contracts. Collect historical risk event data, including pre-risk monitoring data, risk level determination results, and actual risk situations; analyze the relationship between different risk indicators in historical data and actual risk levels; use threshold judgment algorithms to evaluate the performance of current warning thresholds and adjust the warning thresholds for each risk indicator; Use the LSTM network to establish a spatiotemporal series prediction model, analyze historical risk event data, use new risk event data as training samples, input the LSTM model for training, and use cross-validation and regularization techniques to prevent model overfitting. The trained LSTM model is applied to risk warning, and the risk level is predicted based on the current monitoring data, providing a basis for taking preventive measures in advance.

10. A smart construction site construction risk early warning system based on BIM technology, using the smart construction site construction risk early warning method based on BIM technology according to any one of claims 1 to 9, characterized in that: include: Data acquisition and perception module: This includes a multi-source sensor networking unit and a data pre-processing unit. The multi-source sensor networking unit uses multiple sensors to synchronously collect multi-dimensional data from the construction site, ensuring the temporal and spatial consistency of the data. The data pre-processing unit performs filtering, noise reduction, and standardization on the collected raw data. Data Mapping and Analysis Module: This module includes a BIM model mapping unit, a spatiotemporal graph neural network analysis unit, and an autocorrelation analysis unit. The BIM model mapping unit maps pre-processed sensor data to the BIM 3D model to associate the data with the model. The spatiotemporal graph neural network analysis unit uses a spatiotemporal graph neural network to analyze the spatial interactions between people, machinery, and the environment, identifying potential safety risks and abnormal behaviors. The autocorrelation analysis unit performs autocorrelation analysis on historical data to identify periodic risk patterns. Risk value calculation and assessment module: includes a dynamic weight allocation unit, a risk value calculation unit, and a risk level determination unit. The dynamic weight allocation unit combines the analytic hierarchy process (AHP) and the entropy weight method to construct a three-tier indicator system, calculate indicator weights and the comprehensive risk value R. The risk value calculation unit presets risk level classification standards and divides the comprehensive risk value R into low risk, medium risk, and high risk levels. The risk level determination unit determines the final risk level based on the classification standards and the calculated comprehensive risk value. Warning information release and response module: includes a warning information release unit and an emergency plan triggering unit. The warning information release unit releases warning information based on risk levels through BIM model annotation, sound and light alarms, emergency broadcasts, and information push. The emergency plan triggering unit triggers the corresponding emergency plan based on risk levels. Data storage analysis and optimization module: includes blockchain evidence unit, threshold optimization unit and LSTM model training and application unit; among them, the blockchain evidence unit packages the data in the risk warning process into data blocks, adds them to the blockchain in chronological order, and performs recording and verification of risk events; the threshold optimization unit analyzes the relationship between different risk indicators in historical data and the actual risk level, evaluates the performance of the current warning threshold through algorithms, and adjusts the warning threshold of each risk indicator; the LSTM model training and application unit uses the LSTM network to establish a spatiotemporal sequence prediction model, analyzes historical risk event data, trains the LSTM model, and applies the trained LSTM model to real-time risk warning.

Images