Construction safety risk monitoring and early warning method and system based on multi-source perception

Abstract

The invention discloses a construction safety risk monitoring and early warning method and system based on multi-source perception, and relates to the technical field of construction site safety, and the method comprises the steps: constructing a construction site monitoring information network according to a multi-source sensor, and building a construction site three-dimensional construction monitoring space; and constructing a construction safety monitoring model according to the plurality of safety characteristic parameters and the bidirectional long-short-term neural memory network, correcting a risk prediction result through a safety correction factor, and adjusting the corrected risk prediction result in combination with the action track of the constructor. According to the method, construction site data are collected and processed through a multi-source sensor, the processed construction site data are calculated through a bidirectional long-short-term memory network, prediction data are corrected in real time by combining a safety correction factor mechanism and fusing personnel positions and working condition changes, and a four-level grading early warning matrix is constructed; a monitoring-prediction-correction-response closed loop is realized, and the risk early warning response speed is improved.

Description

Technical Field

[0001] This invention relates to the field of construction site safety technology, specifically to a construction safety risk monitoring and early warning method and system based on multi-source sensing. Background Technology

[0002] Construction safety refers to the systematic safeguards implemented at construction sites to prevent personal injury and property damage through technical, managerial, and educational means. Its core is to establish a legally mandated responsibility system, implement dynamic risk management, and integrate physical protection and intelligent monitoring technologies to cover the entire process from commencement to completion, aiming to achieve a fundamental shift from post-event handling to pre-event prevention.

[0003] Current construction safety monitoring is gradually incorporating sensor technology to collect data on personnel location, environmental parameters (such as gas concentration and wind speed), and equipment status (such as tower crane tilt). However, most of these are single-dimensional deployments. While they can identify basic risks, they are difficult to adapt to dynamic changes in the construction environment and scenarios with multiple coupled risks, and cannot meet the needs of refined safety management in complex construction sites. Summary of the Invention

[0004] To address the aforementioned technical issues, this paper provides a construction safety risk monitoring and early warning method and system based on multi-source sensing. This technical solution addresses the problem mentioned in the background that current construction safety monitoring gradually introduces sensor technology, attempting to collect data on personnel location, environmental parameters (such as gas concentration and wind speed), and equipment status (such as tower crane tilt). However, these methods are mostly deployed in a single dimension. Although they can achieve basic risk identification, they are difficult to adapt to the dynamic changes in the construction environment and the coupling of multiple risks, and cannot meet the needs of refined safety management in complex construction sites.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A construction safety risk monitoring and early warning method based on multi-source sensing, comprising: By thoroughly examining the construction site environment and processes, and deploying interconnected multi-source sensors, a construction site monitoring information network is established. Based on the construction site monitoring information network, multi-dimensional information on construction personnel and construction environment is collected, multi-dimensional information datasets are obtained, processed and integrated, and a three-dimensional construction monitoring space for the construction site is established based on the multi-dimensional information datasets. Feature extraction is performed by traversing the three-dimensional construction monitoring space of the construction site to generate multiple safety feature parameters, and a construction safety monitoring model is constructed based on these multiple safety feature parameters. By using a construction safety monitoring model, the on-site environment where construction workers are located is monitored and calculated to obtain risk prediction data; Dynamic analysis is performed on the multidimensional information dataset to generate safety correction factors. Based on the safety correction factors, the risk prediction data is dynamically fed back to generate a trend map of construction environment changes. Based on the trend map of construction environment changes and risk prediction data, construction personnel are monitored in real time, and the risk level is quantified. A multi-level early warning mechanism is established based on the risk level and the location of construction personnel.

[0006] Preferably, the step of traversing the construction site environment and construction process, deploying interconnected multi-source sensors, and constructing a construction site monitoring information network specifically includes: An environmental safety analysis was conducted by thoroughly examining the construction site environment and processes to determine the deployment locations of multi-source sensors. Based on the site environment and construction process, select the type of multi-source sensor, deploy the multi-source sensors according to their locations and associate them with each other to generate a device association chain; Multiple different equipment linkages are linked together and subjected to multidimensional analysis. Based on the analysis results, multiple construction site monitoring sub-networks are divided. Multiple construction site monitoring sub-networks are interconnected to perform data correction and construct a construction site monitoring information network.

[0007] Preferably, the step of collecting multi-dimensional information on construction personnel and the construction environment based on the construction site monitoring information network, obtaining a multi-dimensional information dataset, processing and fusing it, and establishing a three-dimensional construction monitoring space for the construction site based on the multi-dimensional information dataset specifically includes: Based on the construction site monitoring information network, a multi-dimensional analysis of construction personnel and the construction environment is conducted to determine the types of multiple multi-dimensional sensor data. Outliers and missing data in multidimensional sensor data are processed, and the data units and formats of multidimensional data are standardized to generate a clean dataset. Based on the types of cleaning datasets and multidimensional sensor data, multiple construction site monitoring sub-networks are mapped to obtain multiple monitoring spatial maps; Based on multiple monitoring space maps, regression analysis is performed on the types of multidimensional sensor data and the cleaning dataset to generate a monitoring space risk dataset. Based on the real-time analysis of the monitoring space risk dataset, a three-dimensional construction monitoring space for the construction site is constructed.

[0008] Preferably, the step of traversing the three-dimensional construction monitoring space of the construction site to extract features and generate multiple safety feature parameters specifically includes: Based on the type of multidimensional sensor data, the data acquisition time sequence is extracted. The three-dimensional construction monitoring space of the construction site is traversed according to the data acquisition time sequence to conduct risk analysis and confirm the type of construction risk. The three-dimensional construction monitoring space of the construction site is traversed in the order of data collection time to conduct environmental equipment analysis and confirm the status of the site's environmental equipment. The data on construction risk types and site environment and equipment status are standardized to generate multiple safety characteristic parameters.

[0009] Preferably, the construction safety monitoring model based on multiple safety characteristic parameters specifically includes: The security feature parameters are standardized and formatted, and sorted according to the data acquisition time to obtain a security feature parameter sequence table; By integrating the safety characteristic parameter sequence list, construction risk types, and site environment and equipment status from multiple dimensions, a risk status association rule matrix is ​​constructed. The status of the construction site environment and equipment around the construction workers was retrieved, and the training dataset was confirmed based on the risk status association rule matrix and the three-dimensional construction monitoring space of the construction site. The behavioral state of construction workers is introduced, and the training and calibration are performed based on the site environment and equipment status and the training dataset in the order of data collection time. Temporal correlation features in the training dataset are extracted based on the bidirectional long short-term memory network to capture long-short-term dependencies. By fusing temporal correlation features with long-short-term dependencies, dynamically allocating weights according to the data collection time sequence, and iteratively training and calibrating the construction safety monitoring model using a bidirectional long short-term memory network based on the temporal data of the construction site environment and equipment status and the training dataset.

[0010] Preferably, the step of monitoring and calculating the on-site environment where construction workers are located through a construction safety monitoring model to obtain risk prediction data specifically includes: Real-time coordinates of construction workers are extracted from multidimensional information datasets, and their movement trajectories are drawn based on the three-dimensional construction monitoring space. The construction worker status data and construction environment data of the construction worker movement trajectory are filtered, labeled as construction worker surrounding data, and the construction worker surrounding data are standardized to generate a standardized real-time dataset. Risk prediction data is obtained by calculating standardized real-time datasets based on the construction safety monitoring model.

[0011] Preferably, the step of dynamically analyzing the multidimensional information dataset to generate a security correction factor specifically includes: By traversing the multidimensional information dataset, the dimensions of construction personnel behavior, construction environment change, and equipment operation status are obtained, which are referred to as the three core dimensions. Then, time-series dynamic analysis is performed to generate construction personnel behavior risk factors, environmental mutation factors, and equipment impact factors. The safety correction factor is obtained by weighted summation of risk factors for construction workers' behavior, environmental mutation factors, and equipment impact factors. The safety risk contribution value of the three core dimensions is calculated based on the ratio of the safety correction factor to the risk factor of construction personnel behavior, the environmental mutation factor, and the equipment impact factor. A safety correction factor update mechanism is established, which is consistent with the update sampling frequency of multi-source sensors.

[0012] Preferably, the step of dynamically feeding back the risk prediction data based on the safety correction factor to generate a construction environment change trend map specifically includes: Traverse the time-series records of safety correction factors, risk prediction data, and multidimensional information datasets to obtain the core objects that require dynamic correction feedback, and perform time-series dynamic analysis. The risk level quantification value of the core object is globally corrected using the safety correction factor. The probability of occurrence of the core risk type and the safety risk contribution value of the three core dimensions are individually corrected according to the corresponding dimension factor to generate the corrected risk data. Extract the data collection timestamps, sort them in ascending order by time, divide them into fixed time windows, calculate the mean, peak value, and rate of change of the risk data in each window, and generate a time series of time-corrected risk data. The core visualization dimensions of the construction environment change trend map are determined, and the time series data of the three core dimensions are synchronously linked. The time series data of the three core dimensions are visualized and transformed to generate a multi-curve fused construction environment change trend map. At the same time, a dynamic update mechanism for the trend map consistent with the update sampling frequency of the multi-source sensors is established.

[0013] Preferably, the real-time monitoring of construction personnel based on the construction environment change trend map and risk prediction data, the quantification of risk levels, and the establishment of a multi-level early warning mechanism based on the risk level and the location of construction personnel specifically include: By traversing the time-series curve data and risk prediction data in the construction environment change trend chart, the movement trajectory and expected tasks of construction personnel are obtained, and multi-dimensional time-series correlation analysis is performed. By correlating the real-time location and movement trajectory of construction workers with the safety risk contribution values ​​of the three core dimensions, the influence weight of the construction workers' expected tasks on the risk level is analyzed, and the risk evolution slope is combined with the construction environment change trend map to predict the direction of risk change. Based on the quantitative range of the corrected risk prediction data, the risk level threshold is defined, the sampling frequency is updated according to the multi-source sensor, the status data and risk data of construction personnel are continuously collected, the latest curve of the construction environment change trend map is compared with the historical data, the quantitative value of risk level is recalculated, and the risk level label of construction personnel is updated synchronously. Cross-validate the real-time quantified risk level with the safety risk contribution values ​​of the three core dimensions in the construction environment change trend chart, and set thresholds; Based on the risk evolution data of the quantitative risk level value, the real-time location of construction personnel, and the trend map of changes in the construction environment, a multi-factor time-series correlation analysis is conducted to construct a two-dimensional early warning trigger matrix and simultaneously establish an early warning dynamic adjustment mechanism.

[0014] Preferably, a construction safety risk monitoring and early warning system based on multi-source sensing is used to implement the aforementioned construction safety risk monitoring and early warning method based on multi-source sensing, specifically including: The multi-source information acquisition and data processing module is used to analyze the construction site environment and construction process, deploy multi-source sensors to build a monitoring network, collect various types of data such as personnel, environment and equipment, and output high-quality clean fusion datasets after cleaning, standardization and multi-source fusion processing, providing reliable data support for subsequent modules and ensuring data accuracy and consistency. The monitoring space establishment and feature extraction module is used to fuse data to map the monitoring space map, construct a three-dimensional construction monitoring space through regression analysis, traverse the space to extract safety feature parameters, realize the visualization of the construction site environment space, clarify the construction risk type and equipment status, generate targeted safety feature parameters, and support model training. The predictive model training and risk prediction module is used to train a bidirectional long short-term memory network model, predict risk data, generate safety correction factors to correct the data and draw trend charts. The model has accurate risk prediction capabilities, the corrected data is more in line with reality, and the trend charts intuitively present the dynamics of risk, improving the efficiency of risk prediction. The real-time monitoring and risk warning module is used to correlate trend charts with risk data, monitor personnel status in real time, quantify risk levels, establish a multi-level warning mechanism, accurately quantify four levels of risk, provide differentiated and timely warning responses, form a closed-loop management system, and effectively protect the safety of construction personnel.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a monitoring network using multi-source sensors, and establishes a three-dimensional construction monitoring space through data cleaning and fusion, achieving multi-dimensional coverage of personnel, environment, and equipment, thus avoiding monitoring blind spots. A model is built based on a bidirectional long short-term memory network, combined with dynamic calibration of personnel behavior to improve the accuracy of risk prediction; safety correction factors are generated to dynamically correct data, and the sensor update frequency is synchronized to reduce deviations. Furthermore, it can quantify four levels of risk, and establish a multi-level early warning mechanism based on personnel location and risk evolution, ensuring timely and differentiated early warning responses, forming a closed-loop management system, comprehensively protecting construction safety, and significantly improving the efficiency and reliability of monitoring and early warning. Attached Figure Description

[0016] Figure 1 This is a flowchart of a construction safety risk monitoring and early warning method based on multi-source sensing in this invention; Figure 2This is a flowchart of the steps for establishing a three-dimensional construction monitoring space at a construction site in this invention; Figure 3 This is a flowchart of the steps involved in establishing the construction safety monitoring model in this invention; Figure 4 This is a flowchart of the operation steps of a construction safety risk monitoring and early warning system based on multi-source sensing in this invention. Detailed Implementation

[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0018] Reference Figure 1 As shown, a construction safety risk monitoring and early warning method based on multi-source sensing includes: By thoroughly examining the construction site environment and processes, and deploying interconnected multi-source sensors, a construction site monitoring information network is established. Based on the construction site monitoring information network, multi-dimensional information on construction personnel and construction environment is collected, multi-dimensional information datasets are obtained, processed and integrated, and a three-dimensional construction monitoring space for the construction site is established based on the multi-dimensional information datasets. Feature extraction is performed by traversing the three-dimensional construction monitoring space of the construction site to generate multiple safety feature parameters, and a construction safety monitoring model is constructed based on these multiple safety feature parameters. By using a construction safety monitoring model, the on-site environment where construction workers are located is monitored and calculated to obtain risk prediction data; Dynamic analysis is performed on the multidimensional information dataset to generate safety correction factors. Based on the safety correction factors, the risk prediction data is dynamically fed back to generate a trend map of construction environment changes. Based on the trend map of construction environment changes and risk prediction data, construction personnel are monitored in real time, and the risk level is quantified. A multi-level early warning mechanism is established based on the risk level and the location of construction personnel.

[0019] Reference Figure 2 As shown, further, the process of traversing the construction site environment and construction process, deploying interconnected multi-source sensors, and constructing a construction site monitoring information network specifically includes: An environmental safety analysis was conducted by thoroughly examining the construction site environment and processes to determine the deployment locations of multi-source sensors. The construction site environment was surveyed using multiple information sources to create a 3D topographic map. This map marked slope areas with elevation differences ≥2m, the depth and direction of underground pipelines, and distances to nearby buildings. The distribution of obstacles within the site was clarified. The core operational content of each stage of the construction process was analyzed, and the risk evolution patterns at each node were identified. Risk factors (personnel density, equipment operating intensity, environmental impact, and accident probability) for each area were quantitatively scored from 1 to 5 points using expert evaluation and construction safety inspection standards. The deployment locations and monitoring ranges of multi-dimensional sensors were confirmed. A total score ≥12 points was considered excellent. Priority I (mandatory placement), 8-11 points (should be placed), and ≤7 points (optional placement) are ranked as follows: Priority I (mandatory placement), 8-11 points (should be placed), and ≤7 points (optional placement). The placement locations are updated synchronously according to the changes in construction process stages. Example: Deep foundation pit bottom cleaning area: Personnel density is 12 people, area is 12 square meters, score is 4 points; Equipment operation intensity is simultaneous operation of concrete pump truck and small cleaning machinery, score is 4 points; Environmental impact: Deep foundation pit bottom, high-voltage cables within 15m, score is 4 points; Accident probability: High-risk operation, with four types of risks including collapse and burial, score is 4 points. Based on the site environment and construction process, select the type of multi-source sensor, deploy the multi-source sensors according to their locations and associate them with each other to generate a device association chain; Based on the construction stage, site environment, and risk evolution patterns at each node, the types of multi-source sensors are precisely selected, including UWB personnel positioning base stations, tilt sensors, and temperature sensors. At the same time, the multi-source sensors to be deployed at each location are determined according to the location relationship and risk evolution patterns at each node. Multiple multi-source sensors with wired and wireless transmission methods are deployed simultaneously at each location to ensure real-time data availability. Sensors with coordinated functions at different deployment locations within the same area are bound together, and the multi-source sensors are connected using the rules of area coding, sensor type coding, and link sequence number, thereby completing the construction of the equipment association chain. Multiple different equipment linkages are linked together and subjected to multidimensional analysis. Based on the analysis results, multiple construction site monitoring sub-networks are divided. The established equipment association chains are analyzed to identify their core characteristics. Stable connections are established between different equipment association chains through physical gateway interconnection and logical attribute binding rules. Subsequently, the monitoring functions of different association chains are integrated for functional complementarity analysis. By mining the inherent relationships between different association chain data, data from the same source or complementary sources are cross-validated to complete the analysis. Then, association chains with highly overlapping monitoring ranges are identified, and redundant resources are merged to analyze the overlap of risk coverage. Based on the deployment location of the association chains, their spatial coverage continuity is assessed to ensure continuous coverage across the entire construction site, thereby ensuring spatial coverage continuity. The process continues, and the communication protocols of different interconnected chains are verified to be compatible. Interconnected chains that can interact smoothly are grouped into the same sub-network to complete the protocol compatibility check. Then, multi-dimensional analysis is performed to extract the running data of all device interconnected chains during continuous operation and combine them into a cross-chain analysis dataset. The cross-chain dataset is then standardized and sorted according to time order. Following the construction process, the cross-chain dataset is split into sub-datasets for the corresponding stages. Different device interconnected chains are combined according to the sub-datasets to form combined device interconnected chains. At the same time, the combined device interconnected chains are divided into different construction site monitoring sub-networks according to the construction site area. Multiple construction site monitoring sub-networks are interconnected to perform data correction and construct a construction site monitoring information network. With a core gateway, regional gateways, and sub-network gateways as the core, and combined with the core characteristics of the device association chain, a hierarchical interconnection architecture is constructed. All construction site monitoring sub-networks are connected to the sub-network gateways. All construction site monitoring sub-networks are divided according to construction site areas, and the construction site areas are the main access points to the regional gateways. The core gateway verifies and sorts all data in the regional gateways and sub-network gateways, thereby completing the construction of the construction site monitoring information network.

[0020] Furthermore, the process of collecting multi-dimensional information on construction personnel and the construction environment based on the construction site monitoring information network, obtaining a multi-dimensional information dataset, processing and fusing it, and establishing a three-dimensional construction monitoring space for the construction site based on the multi-dimensional information dataset specifically includes: Based on the construction site monitoring information network, a multi-dimensional analysis of construction personnel and the construction environment is conducted to determine the types of multiple multi-dimensional sensor data. Based on multiple construction-related monitoring sub-networks in the construction site monitoring information network, multidimensional data on the movement trajectory of construction personnel and their surrounding construction environment are extracted. The multidimensional data is then unified according to time sequence. Focusing on construction personnel and the construction environment, the multidimensional data is broken down into key analysis dimensions: construction personnel status, construction environment parameters, construction equipment operation, and spatiotemporal correlation. These key analysis dimensions are verified using the construction site monitoring sub-network. Based on the analysis and verification results of these key analysis dimensions, the multidimensional sensor data is further broken down, extracting personnel-related data and environment-equipment-related data. The time sequence of the broken data is unified based on the sampling frequency of multi-source sensors. Finally, with construction personnel safety as the core, a second breakdown is performed according to the construction personnel status dimension, construction environment parameter dimension, construction environment parameter dimension, and construction equipment operation dimension to obtain the types of core multidimensional sensor data. The definition, source, and purpose of these core multidimensional sensor data types are clarified, and a unified label is used to label the types of core multidimensional sensor data. Outliers and missing data in multidimensional sensor data are processed, and the data units and formats of multidimensional data are standardized to generate a clean dataset. Based on industry standards, limit thresholds for the types of core multidimensional sensor data are set. A sliding window strategy is adopted, and the mean value of the data within each window is calculated. If the mean value exceeds the limit threshold, it is marked as a potential anomaly. The mean values ​​of data within multiple adjacent windows are verified simultaneously. If the deviation rate exceeds 20%, it is judged as abnormal data. At the same time, based on industry specifications and construction scenario characteristics, deviation rates between 5% and 20% are judged as mild anomalies, deviation rates between 20% and 50% are judged as moderate anomalies, and deviation rates greater than 50% are judged as severe anomalies. Multi-source sensors are replaced, and data at the deployment locations are re-acquired. Simultaneously, forward padding and the chain of similar sensors and related equipment are used to fill in missing data. Finally, all multidimensional sensor data are standardized and the units of measurement are unified, thereby completing the creation of the cleaning dataset.

[0021] Based on the types of cleaning datasets and multidimensional sensor data, multiple construction site monitoring sub-networks are mapped to obtain multiple monitoring spatial maps; Based on the construction site monitoring subnetwork, and combined with the types of multidimensional sensor data and the characteristics of the cleaning dataset, a mapping rule definition is established: data type corresponds to subnetwork, cleaning data corresponds to data type, and spatial coordinates correspond to subnetwork. According to the mapping rule definition, the cleaning dataset and the types of multidimensional sensor data are associated and matched. The cleaning data generated by multi-source sensors in the cleaning dataset is bound to the subnetwork spatial coordinates of the construction site monitoring subnetwork. Based on the binding relationship between the cleaning dataset and the subnetwork spatial coordinates, a spatial visualization tool is used to generate a monitoring spatial map. Based on multiple monitoring space maps, regression analysis is performed on the types of multidimensional sensor data and the cleaning dataset to generate a monitoring space risk dataset. Real-time analysis is then performed on the monitoring space risk dataset to construct a three-dimensional construction monitoring space for the construction site. Based on the monitoring spatial map, spatial-data-risk correlation attributes are extracted for each map. Cleaning data in the cleaning dataset are grouped according to monitoring spatial map ID and data type to ensure that each data type can be traced back to a specific spatial location. A multivariate linear regression combined with spatial weight correction is used, where the dependent variable is the monitoring spatial risk value and the independent variable is the cleaning monitoring value of the multidimensional sensor data. The correlation between the type of multidimensional sensor data, the cleaning data monitoring value, and the monitoring spatial risk is established, confirming the influence weight of different data types on spatial risk. Risk values ​​are calculated based on the regression model, and a monitoring spatial risk dataset is generated by combining timestamps, spatial coordinates, and risk levels, thus classifying the risk values. The risk level is divided into four levels: Level 1 (low risk): Y∈[0,2); Level 2 (medium risk): Y∈[2,5); Level 3 (high risk): Y∈[5,8); Level 4 (extremely high risk): Y∈[8,10]. A structured monitoring spatial risk dataset is generated. Based on the monitoring spatial risk dataset and the 3D environmental topographic map of the construction site, BIM+GIS fusion technology is used to construct the 3D construction monitoring space of the construction site and establish a coordinate system. Then, according to the sub-network area corresponding to the monitoring spatial map, the spatial blocks are divided in the 3D framework. The sensor coordinates, risk values, and risk levels in the monitoring spatial risk dataset are associated with the 3D spatial blocks to achieve spatial visualization and real-time risk display.

[0022] Reference Figure 3 As shown, further, the step of traversing the three-dimensional construction monitoring space of the construction site to extract features and generate multiple safety feature parameters specifically includes: Based on the type of multidimensional sensor data, the data acquisition time sequence is extracted. The three-dimensional construction monitoring space of the construction site is traversed according to the data acquisition time sequence to conduct risk analysis and confirm the type of construction risk. A time-series list of multidimensional sensor data types was created, categorized by data type, acquisition frequency, and timestamp format, to identify the temporal dimension characteristics of each data type. Subsequently, the 3D construction monitoring space at the site was further subdivided into zones based on region ID, spatial coordinate range, main operation type, and high-risk points. Following industry standards and combining multidimensional sensor data monitoring indicators, a correspondence between data indicators, risk characteristics, and risk types was constructed, forming a risk analysis indicator system. Based on the time-series list of multidimensional sensor data types, each data type was extracted from the cleaning dataset in ascending order of timestamp, and then grouped by monitoring sub-network and time segment. The process involves traversing the 3D construction monitoring space of the site in the order of segment-spatial partition-high-risk point, checking the monitoring data of each spatial location at different time points one by one, and locating potential risk areas (example: according to industry standards: slope areas where personnel are ≤2m from the edge of high-altitude work areas and the excavation depth of foundation pits is ≥3m are directly listed as preset high-risk points, and any abnormality triggered by the monitoring data is judged as a potential risk area). Based on the results of multi-dimensional risk analysis, combined with the correspondence between data indicators, risk characteristics and risk types, annotations are made in the 3D monitoring space to generate a risk type-time-space annotation table, thereby completing the confirmation of risk types. The three-dimensional construction monitoring space of the construction site is traversed in the order of data collection time to conduct environmental equipment analysis and confirm the status of the site's environmental equipment. Based on the equipment configuration of the construction site monitoring information network, a list is established according to equipment type, core monitoring indicators, and associated sensor data types to confirm the status monitoring dimensions of each type of equipment. Then, through the construction site's three-dimensional monitoring space, the physical installation coordinates of each environmental device are precisely bound to its position in the three-dimensional spatial model. A unique icon is set for each type of equipment in the three-dimensional spatial model, and the equipment ID and core monitoring indicators are labeled to ensure that the spatial area where the equipment is located can be located during time-series traversal. Combining equipment operating parameter thresholds and multi-dimensional sensor data characteristics, a correspondence between monitoring indicators, threshold ranges, and equipment status is constructed, classifying equipment status as normal operation. According to industry standards, the standard threshold for allowable deviation of equipment parameters is ±20%. The criteria for minor anomalies are set as parameters exceeding the normal range but ≤20%, with a slow rate of change, indicating that the equipment function basically meets operational requirements and there are no direct safety hazards. According to industry standards, 50% is the critical failure threshold for key safety parameters. The criteria for moderate anomalies are set as parameters exceeding the normal range by 20%-50%, with a faster rate of change, and the equipment function is affected. 50% is the critical failure threshold for key safety parameters. The judgment condition for serious anomalies is set as follows: the parameter exceeds the normal range by ≥50%, or the key safety indicator fails, posing a major safety hazard, and operation must be stopped immediately. Based on the correspondence between the types of environmental equipment-related sensor data, the associated monitoring data of each device is extracted from the cleaning dataset in ascending order of data collection timestamp, and grouped by device ID to ensure that the device data within the same time period can be analyzed in association. According to the time sequence of the device-related data, the three-dimensional construction monitoring space of the construction site is traversed to check the operating data of each environmental device at different time points, locate the time node and spatial location of the abnormal device status, and analyze the monitoring indicator data obtained by the time sequence traversal from three dimensions: stability of operating parameters, functional matching, and coordination of associated devices, based on the equipment status judgment criteria, to identify the abnormal characteristics of the equipment status. Based on the multi-dimensional environmental equipment analysis results, combined with the correspondence between the threshold range and the equipment status, the final status of each time segment and each device is confirmed and visualized in the three-dimensional monitoring space to complete the confirmation of the status of the construction site environmental equipment. The data on construction risk types and site environment and equipment status are standardized to generate multiple safety characteristic parameters; Based on the risk type and equipment status analysis results, core data are extracted from three dimensions: risk quantification indicators, equipment operating parameters, and spatial correlation attributes. The type, unit, and value range of each type of data are confirmed, and the data are standardized to ensure that the processed data range is uniform, mapped to the [0, 1] interval, and the original characteristics and trends of the data are preserved. The qualitative data in the construction risk type and the site environment and equipment status are converted into numerical format. The standardized quantitative data and the coded qualitative data are associated according to time window-spatial location-equipment ID and integrated into structured safety feature parameters.

[0023] Furthermore, the construction safety monitoring model based on multiple safety characteristic parameters specifically includes: The security feature parameters are standardized and formatted, and sorted according to the data acquisition time to obtain a security feature parameter sequence table; Based on the analysis results of construction risk types and equipment status, safety feature parameters cover four core dimensions: risk quantification, equipment operation, personnel correlation, and spatial attributes. Each dimension includes quantitative and qualitative parameters. For different types of feature parameters, the feature parameters are standardized, and the parameter format and naming rules are unified to ensure data structure consistency. For the extracted raw safety feature parameters, the data is first cleaned and then standardized according to the rules to eliminate dimensional differences. The standardized parameters are organized in a unified format, and the formatted single parameter records are arranged in ascending order by timestamp to form a time-series safety feature parameter sequence table. By integrating the safety characteristic parameter sequence list, construction risk types, and site environment and equipment status from multiple dimensions, a risk status association rule matrix is ​​constructed. From the three dimensions of temporal consistency, spatial correlation, and risk coupling, the core attributes of three types of elements are extracted to ensure that data can be aligned and logically correlated during fusion. Fusion rules are formulated from the three dimensions of temporal alignment, spatial matching, and risk coupling, clarifying the association methods and priorities of the three types of elements: Temporal alignment: Based on the timestamp of the safety feature parameter sequence table, construction risk types and equipment status data within the same time window are associated, with a time deviation ≤ 1 second and a priority of 1; Spatial matching: Using the 3D monitoring spatial partition ID + sensor coordinates as keywords, the three types of elements within the same spatial area are associated, with a spatial deviation ≤ 0.5m and a priority of 2; Risk coupling: Based on equipment type The mapping relationship between construction risk types and associated risks is established, with temporal alignment as the core and spatial matching as the constraint. This involves fusing three types of elements to generate a four-dimensional association table. Based on this table, an analytic hierarchy process (AHP) combined with historical data statistics is used to calculate the contribution weight of different equipment types to various risks, quantify the contribution, and map thresholds. The trigger thresholds for construction risk types are linked to equipment status codes, clarifying the quantitative correspondence between equipment status and risk triggering. The coupling relationship between construction risk types and site environment equipment status is quantified. Based on the temporal-spatial fusion results and the risk-equipment coupling quantification relationship, a three-dimensional rule matrix is ​​constructed, thus completing the construction of the risk status association rule matrix. The status of the construction site environment and equipment around the construction workers was retrieved, and the training dataset was confirmed based on the risk status association rule matrix and the three-dimensional construction monitoring space of the construction site. Based on the construction of a 3D construction monitoring space and the risk status association rule matrix, key attributes of four core elements are extracted. Centered on the real-time coordinates of construction personnel, and based on the risk types and risk type threshold mappings related to equipment in the risk status association rule matrix, a circular retrieval range is defined, with the radius adjusted according to equipment type to ensure coverage of potential impact areas. Rules for defining equipment around construction personnel are established. Based on the risk status association rule matrix, only equipment status-risk type data with an association strength ≥ 0.3 are retained as the equipment status-risk association filtering rule. The partitions where construction personnel are located are aligned with those where equipment is located. The real-time location of construction personnel is determined based on UWB positioning data. A list of surrounding equipment whose status needs to be retrieved is selected according to the surrounding equipment definition rules, ensuring a close spatial association between equipment and personnel. Real-time status data of the selected surrounding equipment is retrieved, and the equipment status-risk type association is verified using the risk status association rule matrix. Highly correlated data is filtered, retaining only the correlated data within the partition where construction personnel are located. Furthermore, based on the partition risk level, high-value data with an association strength ≥ 0.4 is prioritized, thus completing the partitioning of the training dataset. The behavioral state of construction workers is introduced, and the training and calibration are performed based on the site environment and equipment status and the training dataset in the order of data collection time. Temporal correlation features in the training dataset are extracted based on the bidirectional long short-term memory network to capture long-short-term dependencies. Based on the 3D construction monitoring space and training dataset, the behavioral states of construction personnel, the status of the construction site environment and equipment, and the training dataset are extracted to formulate three core rules: Classification and coding rules for construction personnel behavioral states: Based on industry standards, behavioral states are divided into four quantifiable core types: Compliant work behavior: Wearing safety helmets, safety belts, protective gloves, and other protective equipment as required; Minor violations: Temporarily removing protective equipment during work breaks; Moderate violations: Not wearing critical protective equipment; Serious violations: Conducting high-risk operations without any protection. A combination of ordered coding and one-hot coding is used to convert the data into a numerical format recognizable by a bidirectional long short-term memory network. Temporal association rules between behavioral states and equipment status: Based on the data collection timestamp, the behavioral states of personnel are correlated with the status of the construction site environment and equipment in a 1:1 temporal ratio. The timing alignment must be ≤1 second; the timing order rules for training and calibration are as follows: training and calibration are performed in ascending order of data acquisition time to ensure that the Bi-LSTM model can learn the evolution law of behavior-equipment-risk according to the real time series; then, the behavioral status data of construction personnel is obtained through multi-source perception fusion, and is cleaned, encoded and standardized to ensure that the data quality meets the model training requirements. The preprocessed behavioral status data of construction personnel is fused with the confirmed site environment equipment status and training dataset in three dimensions according to timestamps and spatial coordinates to construct a complete training sample, ensuring that the temporal logic is consistent with the risk evolution law of the construction scenario. The calibrated training dataset is used through the forward LSTM + backward LSTM dual-link structure of the Bi-LSTM model to extract temporal correlation features and capture long-short-term dependencies. By fusing temporal correlation features with long-short-term dependencies, dynamically allocating weights according to the data collection time sequence, and iteratively training and calibrating a construction safety monitoring model based on the temporal data of the construction site environment and equipment status and the training dataset through a bidirectional long short-term memory network; Core rules are formulated by extracting the core elements of temporal correlation features and long-term / short-term dependencies. These rules include: a fusion rule for temporal correlation features and long-term / short-term dependencies using a weighted fusion method of feature value × dependency attention weight to strengthen the synergistic effect of highly correlated features and strong dependencies; a dynamic weight allocation rule based on time order, dynamically adjusting feature weights according to the evolution of construction risks and the time sequence of data collection; and the core logic of feature weight adjustment, including weight tilting during high-risk periods, weight adjustment during periods of risk accumulation, and weight normalization constraints. Subsequently, based on the fusion rules of temporal correlation features and long-term / short-term dependencies, the features extracted by Bi-LSTM are... The time-series correlation features and long-term and short-term dependencies are weighted and fused to generate a comprehensive feature vector. According to the dynamic weight allocation rule of time order, the weights of the fused features are adjusted according to time order and risk status to ensure that the model prioritizes high-risk periods and highly correlated features. Then, the fused feature vector is combined with the dynamic weights and input into the Bi-LSTM model for iterative training. After each training round, the model is calibrated according to the deviation to gradually improve the prediction accuracy. Based on the iterative training and calibration results, the final structure and parameters of the construction safety monitoring model are confirmed, and the performance is evaluated from three dimensions: prediction accuracy, generalization ability, and real-time performance, thus completing the construction of the construction safety monitoring model.

[0024] Furthermore, the monitoring and calculation of the on-site environment where construction workers are located through the construction safety monitoring model to obtain risk prediction data specifically includes: Real-time coordinates of construction workers are extracted from multidimensional information datasets, and their movement trajectories are drawn based on the three-dimensional construction monitoring space. Based on the construction of the 3D construction monitoring space and the processing of multidimensional information datasets, three core rules were formulated: real-time coordinate extraction rules (priority of data sources, determination of coordinate validity, and alignment of timestamps); 3D spatial coordinate mapping rules (coordinate system unification and partition matching); and movement trajectory drawing rules (consistent trajectory sampling density, clear trajectory visualization attributes, and clear labeling of risk areas). Subsequently, the positioning data of construction personnel were selected from the multidimensional information dataset, and real-time coordinates were extracted and verified according to the three core rules. The extracted geodetic coordinates were matched with the partitions and reference points of the 3D construction monitoring space to ensure that the coordinates fell within a reasonable spatial range. Then, based on the real-time coordinates after mapping and verification, the movement trajectory was drawn in the 3D construction monitoring space model, and visualization optimization and risk association labeling were performed. The construction worker status data and construction environment data of the construction worker movement trajectory are filtered, labeled as construction worker surrounding data, and the construction worker surrounding data are standardized to generate a standardized real-time dataset. Based on multidimensional information datasets, 3D construction monitoring space, and construction personnel movement trajectories, core data is extracted from three dimensions: personnel status, construction environment, and spatial correlation. Surrounding data definition rules are established: a circular filtering range is defined centered on the real-time coordinates of construction personnel, with the radius dynamically adjusted according to data type; data annotation rules: the filtered personnel status data and construction environment data are structured and annotated according to data category, risk correlation, and collection time; standardization processing rules: differentiated standardization methods are adopted for different types of data; based on the construction personnel movement trajectories, the corresponding time-series coordinates are extracted, and associated personnel status data and construction environment data are filtered according to spatial range and time synchronization rules. The filtered personnel status data, construction environment data, and spatial correlation data are integrated according to unified annotation rules to form a structured construction personnel surrounding dataset. The annotated construction personnel surrounding dataset is then standardized according to quantitative and qualitative data categories to eliminate dimensional differences, ultimately integrating into a structured standardized real-time dataset. Risk prediction data is obtained by calculating standardized real-time datasets based on the construction safety monitoring model; Based on the construction of the construction safety monitoring model and the generation of standardized real-time datasets, key attributes of three core elements are extracted, the matching conditions between data input and model calculation are clarified, and data input adaptation rules are formulated: time series length, feature dimension matching, and timestamp synchronization; model calculation logic rules: feature weight invocation, long-term and short-term dependency calculation, and risk quantification output; prediction result verification rules: confidence threshold, deviation rate constraint, and outlier handling; the standardized real-time dataset is preprocessed, and the adapted time series is input into the construction safety monitoring model to complete the prediction. The prediction results output by the model are verified in multiple dimensions, and data with low confidence and excessive deviation are corrected. The verified and corrected prediction results are then structured and integrated to generate risk prediction data containing information.

[0025] Furthermore, the step of dynamically analyzing the multidimensional information dataset to generate a security correction factor specifically includes: By traversing the multidimensional information dataset, the dimensions of construction personnel behavior, construction environment change, and equipment operation status are obtained, which are referred to as the three core dimensions. Then, time-series dynamic analysis is performed to generate construction personnel behavior risk factors, environmental mutation factors, and equipment impact factors. Based on the construction of a multidimensional information dataset and the three-dimensional construction monitoring space of the construction site, the multidimensional information dataset (extracting data types, timestamps, monitoring indicators, and data status), three core dimensions (extracting dimension names, core monitoring indicators, data sources, and time series analysis cycles), and risk factors (extracting factor names, quantification ranges, calculation basis, and associated risk types) are extracted. Dataset traversal rules (traversal range, data filtering logic, and traversal order), time series dynamic analysis rules (time window division, trend analysis methods, and data standardization), and risk factor generation rules (quantification logic, value constraints, and associated risk types) are formulated. The multidimensional information dataset is sorted according to the traversal rules to form a structured subset of dimensional data. A combination of fixed window trend analysis and sliding window mutation detection is used for processing to mine time series characteristics, identify risk evolution patterns, and calculate construction worker behavior risk factors, environmental mutation factors, and equipment impact factors based on the time series dynamic analysis results, according to the quantification logic of anomaly degree × duration × impact range. The safety correction factor is obtained by weighted summation of risk factors for construction workers' behavior, environmental mutation factors, and equipment impact factors. The risk factors of construction worker behavior (denoted as) ), environmental mutation factors (denoted as ), Equipment Impact Factor (denoted as The risk of a construction worker's behavior, environmental change, and equipment impact factors are mapped to the [0,10] range through time-series dynamic analysis. Higher values ​​indicate higher risks in the corresponding dimension. These factors are collectively referred to as dimension factors. Based on historical accident data from construction sites and building safety inspection standards, an expert evaluation method is used to assess the weight allocation. Specifically, the construction worker behavior dimension is assigned the highest weight (0.45) according to building safety inspection standards and the technical specifications for high-altitude operations in building construction. The environmental change dimension, due to its susceptibility to group risks caused by sudden environmental changes, has the second highest weight (0.30). The equipment operating status dimension, where equipment failures are mostly localized risks and their probability can be reduced through regular maintenance, has a relatively low weight (0.25). A weighted summation calculation is then performed to determine the safety correction factor. The weighted summation formula is: ;in, These are the weight coefficients for each factor. (This is the verified dimension factor), and then the security correction factor is obtained (Example: known). =7.2 (Construction workers walking illegally on the edge of the scaffolding) =4.5 (localized sudden level 5 gusts) =5.8 (Minor fault in construction elevator door interlocking device) The calculated safety correction factor is: ); The safety risk contribution value of the three core dimensions is calculated based on the ratio of the safety correction factor to the risk factor of construction personnel behavior, the environmental mutation factor, and the equipment impact factor. A safety correction factor update mechanism is established, which is consistent with the update sampling frequency of multi-source sensors. The safety risk contribution value of the three core dimensions is calculated based on the ratio of the single-dimensional factor weighted value to the safety correction factor: Dimensional safety risk contribution value = Weighted value of the dimension factor ÷ Safety correction factor; where, the dimension factor weighted value = Dimension factor × corresponding weight coefficient; Core specifications for establishing the update mechanism: Sampling frequency synchronization: Multi-source sensors uniformly collect data once per second, and the safety correction factor is recalculated once every second; Data source priority: Real-time data from the clean dataset is used first. If the clean dataset is not updated, the original sensor data is temporarily called and corrected; Synchronous setting of update trigger conditions: Timed trigger every second or any dimension factor.

[0026] Furthermore, the step of dynamically feeding back the risk prediction data based on the safety correction factor to generate a trend map of changes in the construction environment specifically includes: Traverse the time-series records of safety correction factors, risk prediction data, and multidimensional information datasets to obtain the core objects that require dynamic correction feedback, and perform time-series dynamic analysis. Based on the time-series records and linked analysis of three types of data, the core object types are obtained through risk level screening, safety correction factor sensitivity screening, and data correlation screening. A unique identifier is assigned to each core object. Core objects requiring dynamic correction are selected from three core dimensions: personnel, environment, and equipment. Dynamic analysis is conducted from three dimensions: time-series trend, related impact, and abnormal mutations to confirm the correction direction and priority. Based on the time-series dynamic analysis results, the intensity of correction needs for core objects is quantitatively assessed and sorted by priority: a correction need intensity scoring system is set up, and correction priorities are divided according to the scoring results. For example, a hot work area has a correction need score of 9 points (4 points for a trend slope of 0.03 / second, 3 points for 4 strongly correlated factors, and 3 points for abnormal gas concentration mutations), and is determined to be a first-level priority, triggering emergency correction; a logistics area personnel has a correction need score of 3 points (0 points for a stable trend, 1 point for 1 strongly correlated factor, and 2 points for no abnormal mutations), and is determined to be a third-level priority, included in batch correction. The risk level quantification value of the core object is globally corrected using the safety correction factor. The probability of occurrence of the core risk type and the safety risk contribution value of the three core dimensions are individually corrected according to the corresponding dimension factor to generate the corrected risk data. Using a safety correction factor as the core, the risk level quantification value of the core object is comprehensively corrected. The correction formula is: Corrected risk level quantification value = Original risk level quantification value × Global safety correction factor. For the core object, the probability of occurrence of the core risk type and the safety risk contribution value of the three core dimensions are individually corrected according to the corresponding dimension sub-factors to improve the targeting of the correction. The globally corrected risk level quantification value and the individually corrected core risk probability are combined, where the corrected probability of occurrence = Original probability of occurrence × Corresponding dimension sub-factor; the three dimension contribution values ​​are integrated: Corrected contribution value = Original contribution value × Corresponding dimension sub-factor; after correction, normalization is performed to form complete corrected risk data, which is then output as the corrected risk data. Extract the data collection timestamps, sort them in ascending order by time, divide them into fixed time windows, calculate the mean, peak value, and rate of change of the risk data in each window, and generate a time series of time-corrected risk data. Complete timestamp information is extracted and standardized to ensure consistency in the time dimension. All corrected risk data is traversed, and the collection timestamp of each data point is extracted to ensure consistency with the original data collection time of the multi-source sensors. Based on the risk evolution speed and data collection frequency of the construction scenario, fixed time windows are set differently: a short window with a duration of 1 minute is used for first-priority core objects to adapt to the rapid changes in high-risk scenarios; a medium window with a duration of 5 minutes is used for second-priority core objects to balance accuracy and efficiency; and a long window with a duration of 10 minutes is used for third-priority core objects to adapt to low-fluctuation scenarios. For each fixed time window, the mean, peak value, and rate of change of the risk data are calculated. All three-dimensional indicators of all windows are arranged in ascending order according to the window start time, associated with the core object information, and a time series sequence of time-corrected risk data is generated. The core visualization dimensions of the construction environment change trend map are determined, the time series data of the three core dimensions are synchronously associated, the time series data of the three core dimensions are visualized and transformed to generate a multi-curve fused construction environment change trend map, and a dynamic update mechanism for the trend map consistent with the update sampling frequency of the multi-source sensors is established simultaneously. Combining the core needs of risk monitoring with the practicality of engineering decision-making, four core visualization dimensions are identified: Core Risk Level Dimension: Based on the time series of the corrected risk level quantification values, the overall risk level changes are visually presented using risk level color coding: Level I Low Risk: Green; Level II Moderate Risk: Blue; Level III Relatively High Risk: Yellow; Level IV Extremely High Risk: Red; The time series data of the corrected risk contribution values ​​of the three core dimensions of personnel, environment, and equipment are used as sub-dimensions, each corresponding to a visualization curve; Peak data in the time series are extracted and marked on the trend chart with red dots and floating prompt boxes. The system displays the specific peak value, core risk type, occurrence time stamp, and contribution value of related dimensions. For the risk change rate dimension, time-series data is converted into trend line slope indicators (upward trend: red arrow pointing upwards, stable trend: black horizontal line, downward trend: green arrow pointing downwards), marked at the end of the corresponding time window to intuitively reflect the speed of risk evolution. Using the time axis as a unified benchmark and linking it with time-series data, the linked time-series data is transformed into a multi-curve fusion construction environment change trend chart. The trend chart update frequency is synchronously set according to the multi-source sensor update sampling frequency to ensure that the visualization results reflect the latest risk dynamics in real time.

[0027] Furthermore, the method involves real-time monitoring of construction personnel based on the construction environment change trend map and risk prediction data, quantifying the risk level, and establishing a multi-level early warning mechanism based on the risk level and the location of construction personnel. Specifically, this includes: By traversing the time-series curve data and risk prediction data in the construction environment change trend chart, the movement trajectory and expected tasks of construction personnel are obtained, and multi-dimensional time-series correlation analysis is performed. By integrating time-series curve data from the construction environment change trend map with full-volume risk prediction data and expected tasks, the core characteristics of the analysis object are clarified. Using the time axis as a benchmark, time-series correlation analysis is conducted from trajectory-risk linkage / task-risk matching and dimension-task association. The results of multi-dimensional time-series correlation analysis are quantified into indicators and labels to form a structured analysis report. Trajectory-risk linkage compares the synchronicity of changes in the construction personnel's movement trajectory and the risk time-series curve, calculating the time difference between key trajectory points and risk level changes. Task-risk matching compares the fit between expected tasks and core risk types, matching the preset risk types corresponding to tasks with the actual monitored core risk types. Dimension-task association analyzes the correspondence between the contribution values ​​of the three core dimensions and expected tasks, clarifying the impact weight of different tasks on the risks of each dimension. By correlating the real-time location and movement trajectory of construction workers with the safety risk contribution values ​​of the three core dimensions, the influence weight of the construction workers' expected tasks on the risk level is analyzed, and the risk evolution slope is combined with the construction environment change trend map to predict the direction of risk change. Using the construction worker ID as the core index, real-time location, movement trajectory, and safety risk contribution values ​​of three core dimensions are precisely linked. This binds a complete data chain of real-time location, movement trajectory, risk contribution value, and expected task for each construction worker. Subsequently, a task type-risk type mapping table is established. The preset risk types corresponding to the expected tasks are compared with the core risk types actually monitored. A dual-track method of calculating matching degree and contribution value attribution is used to quantify the impact weight of the construction worker's expected tasks on the risk level: the impact weight (W) of the task on the risk level is calculated as W = (matching degree × 0.6) + (mean contribution value of dominant dimension × 0.4). Then, based on the time-series curve data of the construction environment change trend map, the slope of risk evolution is calculated to quantify the speed and intensity of risk change. By superimposing regional risk attributes and personnel trajectory characteristics, the direction of risk change is comprehensively predicted, and clear prediction conclusions and risk level prompts are output. Among them, the matching degree is calculated as the number of matched risk types / the total number of preset risk types × 100%, and the contribution value attribution is calculated by extracting the time-series data of the contribution values ​​of the three core dimensions during the execution of the expected task and calculating the mean and percentage fluctuation of the contribution value of each dimension. Based on the quantitative range of the corrected risk prediction data, the risk level threshold is defined, the sampling frequency is updated according to the multi-source sensor, the status data and risk data of construction personnel are continuously collected, the latest curve of the construction environment change trend map is compared with the historical data, the quantitative value of risk level is recalculated, and the risk level label of construction personnel is updated synchronously. Based on the reasonable range of the revised risk level quantification values, and referring to industry safety standards, historical accident data, and the risk characteristics of the construction scenario, four risk level thresholds are defined: Level 1 Low Risk: Quantification value 0-0.3, corresponding to a safe and controllable state, no special control requirements, maintain routine monitoring; Level 2 General Risk: Quantification value 0.3-0.6, corresponding to a basically controllable state, requiring strengthened reminders to personnel on standardized operating procedures, and data review every 5 minutes; Level 3 Higher Risk: Quantification value 0.6-0.8, corresponding to a risk warning state, immediately push notifications to workers and on-site safety officers, suspend high-risk operations; Level 4 Extremely High Risk: Quantification value 0.8-1.0, corresponding to an emergency control state. Triggering audible and visual alarms and emergency response linkage requires personnel to immediately evacuate to a safe area. Simultaneously, considering changes in the construction process, new work types, and historical risk data statistics, the rationality of the threshold is verified. Based on the updated sampling frequency of multi-source sensors, real-time data on the status of construction personnel and risk data is collected. Based on the real-time collected data and historical data from the construction environment change trend chart, the risk level quantification value is recalculated and compared with historical data curves for the same work type and area. The analysis focuses on differences in trend slope, frequency of peak occurrences, and changes in the proportion of dimensional contribution values. A composite calculation model combining real-time data weighting and historical trend correction is adopted, with the formula: Recalculated risk level quantification value (R'') = ×Real-time risk data weighting+ × Historical trend correction factor, where The weight for real-time data is set to 0.7. The historical trend weight is set to 0.3. Based on the recalculated risk level quantification value, the corresponding risk level threshold is matched, and the risk level label of the construction personnel is updated in real time. Cross-validate the real-time quantified risk level with the safety risk contribution values ​​of the three core dimensions in the construction environment change trend chart, and set thresholds; Based on the core logic that overall risk is the superposition of risks across various dimensions, a dual verification relationship is established: consistency verification between the real-time risk level and the sum of the contribution values ​​of each dimension; and correlation verification between the contribution value ratio of each dimension and the corresponding risk type. Cross-validation is conducted through a dual path of consistency verification and correlation verification. Consistency verification is performed by calculating the deviation rate between the real-time risk level and the sum of the contribution values ​​of the three dimensions, while correlation verification is performed based on the matching of contribution value ratios with risk types. Abnormal data is eliminated, and combined with the valid data that has passed the verification, a dual threshold system is defined according to the correlation pattern between risk level and dimension contribution value. The dimension contribution value threshold is set in conjunction with the risk level. Furthermore, a threshold linkage rule is established to bind the risk level threshold and the dimension contribution value threshold together, and a normalized optimization mechanism is established: 10% of the valid data samples are sampled weekly, and cross-validation is carried out again to calculate the verification pass rate. If the pass rate is <90%, the reasons are analyzed, and the threshold range or verification formula is adjusted. Based on the risk evolution data of the risk level quantification value, the real-time location of construction personnel, and the trend map of changes in the construction environment, a multi-factor time-series correlation analysis is conducted to construct a two-dimensional early warning trigger matrix and simultaneously establish an early warning dynamic adjustment mechanism. Using the timeline as the core, the system links risk level quantification values, real-time locations of construction personnel, and risk evolution data to uncover temporal linkage patterns among these elements. A two-dimensional early warning trigger matrix is ​​constructed with risk level quantification values ​​on the vertical axis and personnel location-evolution slope combination on the horizontal axis. This matrix identifies the early warning levels and trigger rules corresponding to different combinations. In conjunction with changes in construction scenarios and data feedback, a threshold adjustment mechanism is established: the early warning accuracy rate is statistically analyzed quarterly based on historical early warning data (target ≥ 90%). If the false alarm rate of a certain combination is ≥ 20%, the corresponding early warning level is adjusted. During the transition between construction phases, a dynamic optimization mechanism for updating regional risk attributes and level calibration is implemented to ensure that the early warning matrix continuously aligns with the actual situation on site.

[0028] Reference Figure 4 As shown, further, a construction safety risk monitoring and early warning system based on multi-source sensing is provided to implement a construction safety risk monitoring and early warning method based on multi-source sensing, characterized in that it includes: The multi-source information acquisition and data processing module is used to analyze the construction site environment and construction process, deploy multi-source sensors to build a monitoring network, collect various types of data such as personnel, environment and equipment, and output high-quality clean fusion datasets after cleaning, standardization and multi-source fusion processing, providing reliable data support for subsequent modules and ensuring data accuracy and consistency. The monitoring space establishment and feature extraction module is used to fuse data to map the monitoring space map, construct a three-dimensional construction monitoring space through regression analysis, traverse the space to extract safety feature parameters, realize the visualization of the construction site environment space, clarify the construction risk type and equipment status, generate targeted safety feature parameters, and support model training. The predictive model training and risk prediction module is used to train a bidirectional long short-term memory network model, predict risk data, generate safety correction factors to correct the data and draw trend charts. The model has accurate risk prediction capabilities, the corrected data is more in line with reality, and the trend charts intuitively present the dynamics of risk, improving the efficiency of risk prediction. The real-time monitoring and risk warning module is used to correlate trend charts with risk data, monitor personnel status in real time, quantify risk levels, establish a multi-level warning mechanism, accurately quantify four levels of risk, provide differentiated and timely warning responses, form a closed-loop management system, and effectively protect the safety of construction personnel.

[0029] The advantages of this invention are as follows: A multi-source sensor network is constructed by traversing the construction site environment and processes to build a monitoring information network; multi-dimensional data is collected, processed, and fused to establish a three-dimensional construction monitoring space; then, safety characteristic parameters are extracted, and a dynamically trained and calibrated construction safety monitoring model is built using a bidirectional long short-term memory network; risk prediction data is calculated using this model; then, safety correction factors are generated through dynamic analysis of multi-dimensional data to correct the risk prediction data and generate a trend chart of construction environment changes; finally, the trend chart and risk prediction data are combined to monitor construction personnel in real time, quantify risk levels, and establish a multi-level early warning mechanism, achieving comprehensive and accurate monitoring, scientific and efficient prediction, dynamic data correction, and timely and graded early warnings, effectively ensuring construction safety.

[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A construction safety risk monitoring and early warning method based on multi-source sensing, characterized in that, include: By thoroughly examining the construction site environment and processes, and deploying interconnected multi-source sensors, a construction site monitoring information network is established. Based on the construction site monitoring information network, multi-dimensional information on construction personnel and construction environment is collected, multi-dimensional information datasets are obtained, processed and integrated, and a three-dimensional construction monitoring space for the construction site is established based on the multi-dimensional information datasets. Feature extraction is performed by traversing the three-dimensional construction monitoring space of the construction site to generate multiple safety feature parameters, and a construction safety monitoring model is constructed based on the multiple safety feature parameters. By using a construction safety monitoring model, the on-site environment where construction workers are located is monitored and calculated to obtain risk prediction data; Dynamic analysis is performed on the multidimensional information dataset to generate safety correction factors. Based on the safety correction factors, the risk prediction data is dynamically fed back to generate a trend map of construction environment changes. Based on the trend map of construction environment changes and risk prediction data, construction personnel are monitored in real time, and the risk level is quantified. A multi-level early warning mechanism is established based on the risk level and the location of construction personnel.

2. The construction safety risk monitoring and early warning method based on multi-source sensing according to claim 1, characterized in that, The process of traversing the construction site environment and construction process, deploying interconnected multi-source sensors, and constructing a construction site monitoring information network specifically includes: An environmental safety analysis was conducted by thoroughly examining the construction site environment and processes to determine the deployment locations of multi-source sensors. Based on the site environment and construction process, select the type of multi-source sensor, deploy the multi-source sensors according to their locations and associate them with each other to generate a device association chain; Multiple different equipment linkages are linked together and subjected to multidimensional analysis. Based on the analysis results, multiple construction site monitoring sub-networks are divided. Multiple construction site monitoring sub-networks are interconnected to perform data correction and construct a construction site monitoring information network.

3. The construction safety risk monitoring and early warning method based on multi-source sensing according to claim 1, characterized in that, The process involves collecting multi-dimensional information on construction personnel and the construction environment based on the construction site monitoring information network, obtaining a multi-dimensional information dataset, processing and fusing it, and establishing a three-dimensional construction monitoring space for the construction site based on the multi-dimensional information dataset. Specifically, this includes: Based on the construction site monitoring information network, a multi-dimensional analysis of construction personnel and the construction environment is conducted to determine the types of multiple multi-dimensional sensor data. Outliers and missing data in multidimensional sensor data are processed, and the data units and formats of multidimensional data are standardized to generate a clean dataset. Based on the types of cleaning datasets and multidimensional sensor data, multiple construction site monitoring sub-networks are mapped to obtain multiple monitoring spatial maps; Based on multiple monitoring space maps, regression analysis is performed on the types of multidimensional sensor data and the cleaning dataset to generate a monitoring space risk dataset. Based on the real-time analysis of the monitoring space risk dataset, a three-dimensional construction monitoring space for the construction site is constructed.

4. The construction safety risk monitoring and early warning method based on multi-source sensing according to claim 3, characterized in that, The process involves traversing the three-dimensional construction monitoring space of the construction site to extract features and generate multiple safety feature parameters, specifically including: Based on the type of multidimensional sensor data, the data acquisition time sequence is extracted. The three-dimensional construction monitoring space of the construction site is traversed according to the data acquisition time sequence to conduct risk analysis and confirm the type of construction risk. The three-dimensional construction monitoring space of the construction site is traversed in the order of data collection time to conduct environmental equipment analysis and confirm the status of the site's environmental equipment. The data on construction risk types and site environment and equipment status are standardized to generate multiple safety characteristic parameters.

5. The construction safety risk monitoring and early warning method based on multi-source sensing according to claim 4, characterized in that, The construction safety monitoring model based on multiple safety characteristic parameters specifically includes: The security feature parameters are standardized and formatted, and sorted according to the data acquisition time to obtain a security feature parameter sequence table; By integrating the safety characteristic parameter sequence list, construction risk types, and site environment and equipment status from multiple dimensions, a risk status association rule matrix is ​​constructed. The status of the construction site environment and equipment around the construction workers was retrieved, and the training dataset was confirmed based on the risk status association rule matrix and the three-dimensional construction monitoring space of the construction site. The behavioral state of construction workers is introduced, and the training and calibration are performed based on the site environment and equipment status and the training dataset in the order of data collection time. Temporal correlation features in the training dataset are extracted based on the bidirectional long short-term memory network to capture long-short-term dependencies. By fusing temporal correlation features with long-short-term dependencies, dynamically allocating weights according to the data collection time sequence, and iteratively training and calibrating the construction safety monitoring model using a bidirectional long short-term memory network based on the temporal data of the construction site environment and equipment status and the training dataset.

6. The construction safety risk monitoring and early warning method based on multi-source sensing according to claim 5, characterized in that, The process of monitoring and calculating the on-site environment where construction workers are located through a construction safety monitoring model to obtain risk prediction data specifically includes: Real-time coordinates of construction workers are extracted from multidimensional information datasets, and their movement trajectories are drawn based on the three-dimensional construction monitoring space. The construction worker status data and construction environment data of the construction worker movement trajectory are filtered, labeled as construction worker surrounding data, and the construction worker surrounding data are standardized to generate a standardized real-time dataset. Risk prediction data is obtained by calculating standardized real-time datasets based on the construction safety monitoring model.

7. The construction safety risk monitoring and early warning method based on multi-source sensing according to claim 6, characterized in that, The step of dynamically analyzing a multidimensional information dataset to generate a security correction factor specifically includes: By traversing the multidimensional information dataset, the dimensions of construction personnel behavior, construction environment change, and equipment operation status are obtained, which are referred to as the three core dimensions. Then, time-series dynamic analysis is performed to generate construction personnel behavior risk factors, environmental mutation factors, and equipment impact factors. The safety correction factor is obtained by weighted summation of risk factors for construction workers' behavior, environmental mutation factors, and equipment impact factors. The safety risk contribution value of the three core dimensions is calculated based on the ratio of the safety correction factor to the risk factor of construction personnel behavior, the environmental mutation factor, and the equipment impact factor. A safety correction factor update mechanism is established, which is consistent with the update sampling frequency of multi-source sensors.

8. The construction safety risk monitoring and early warning method based on multi-source sensing according to claim 7, characterized in that, The step of dynamically feeding back risk prediction data based on safety correction factors to generate a construction environment change trend map specifically includes: Traverse the time-series records of safety correction factors, risk prediction data, and multidimensional information datasets to obtain the core objects that require dynamic correction feedback, and perform time-series dynamic analysis. The risk level quantification value of the core object is globally corrected using the safety correction factor. The probability of occurrence of the core risk type and the safety risk contribution value of the three core dimensions are individually corrected according to the corresponding dimension factor to generate the corrected risk data. Extract the data collection timestamps, sort them in ascending order by time, divide them into fixed time windows, calculate the mean, peak value, and rate of change of the risk data in each window, and generate a time series of time-corrected risk data. The core visualization dimensions of the construction environment change trend map are determined, and the time series data of the three core dimensions are synchronously linked. The time series data of the three core dimensions are visualized and transformed to generate a multi-curve fused construction environment change trend map. At the same time, a dynamic update mechanism for the trend map consistent with the update sampling frequency of the multi-source sensors is established.

9. A construction safety risk monitoring and early warning method based on multi-source sensing according to claim 8, characterized in that, The method involves real-time monitoring of construction personnel based on construction environment change trend maps and risk prediction data, quantifying risk levels, and establishing a multi-level early warning mechanism based on risk levels and the location of construction personnel. Specifically, this includes: By traversing the time-series curve data and risk prediction data in the construction environment change trend chart, the movement trajectory and expected tasks of construction personnel are obtained, and multi-dimensional time-series correlation analysis is performed. By correlating the real-time location and movement trajectory of construction workers with the safety risk contribution values ​​of the three core dimensions, the influence weight of the construction workers' expected tasks on the risk level is analyzed, and the risk evolution slope is combined with the construction environment change trend map to predict the direction of risk change. Based on the quantitative range of the corrected risk prediction data, the risk level threshold is defined, the sampling frequency is updated according to the multi-source sensor, the status data and risk data of construction personnel are continuously collected, the latest curve of the construction environment change trend map is compared with the historical data, the quantitative value of risk level is recalculated, and the risk level label of construction personnel is updated synchronously. Cross-validate the real-time quantified risk level with the safety risk contribution values ​​of the three core dimensions in the construction environment change trend chart, and set thresholds; Based on the risk evolution data of the quantitative risk level value, the real-time location of construction personnel, and the trend map of changes in the construction environment, a multi-factor time-series correlation analysis is conducted to construct a two-dimensional early warning trigger matrix and simultaneously establish an early warning dynamic adjustment mechanism.

10. A construction safety risk monitoring and early warning system based on multi-source sensing for implementing the construction safety risk monitoring and early warning method based on multi-source sensing as described in claims 1-9, characterized in that, include: The multi-source information acquisition and data processing module is used to analyze the construction site environment and construction process, deploy multi-source sensors to build a monitoring network, collect various types of data such as personnel, environment and equipment, and output high-quality clean fusion datasets after cleaning, standardization and multi-source fusion processing, providing reliable data support for subsequent modules and ensuring data accuracy and consistency. The monitoring space establishment and feature extraction module is used to fuse data to map the monitoring space map, construct a three-dimensional construction monitoring space through regression analysis, traverse the space to extract safety feature parameters, realize the visualization of the construction site environment space, clarify the construction risk type and equipment status, generate targeted safety feature parameters, and support model training. The predictive model training and risk prediction module is used to train a bidirectional long short-term memory network model, predict risk data, generate safety correction factors to correct the data and draw trend charts. The model has accurate risk prediction capabilities, the corrected data is more in line with reality, and the trend charts intuitively present the dynamics of risk, improving the efficiency of risk prediction. The real-time monitoring and risk warning module is used to correlate trend charts with risk data, monitor personnel status in real time, quantify risk levels, establish a multi-level warning mechanism, accurately quantify four levels of risk, provide differentiated and timely warning responses, form a closed-loop management system, and effectively protect the safety of construction personnel.

Images