Risk prediction method and system for building construction

Abstract

The invention relates to the field of building construction safety, in particular to a risk prediction method and system for building construction. Aiming at the defects of multi-source data isolated analysis, dynamic risk response lagging, insufficient prediction precision and the like in the prior art, a unified analysis base is formed by constructing a space-time fusion data space and integrating multi-dimensional dynamic data such as structure micro-deformation monitoring, environmental parameters, three-dimensional live-action scanning, personnel positioning, a building information model and the like; based on a deep neural network architecture, designing a multi-modal feature extraction mechanism to quantify the coupling risk, and generating a partition risk probability distribution diagram; and in combination with a construction stage characteristic matching security policy library, implementing a three-level early warning mechanism and an automatic avoidance instruction. A closed-loop optimization mechanism is introduced, model parameters and decision threshold values are dynamically adjusted through actual accident feedback, and continuous evolution of a prediction system is achieved. According to the method, the active prevention and control capacity of compound accidents such as collapse and high-altitude falling is remarkably improved, and a self-adaptive intelligent protection system is constructed for a construction site.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of construction safety, and particularly relates to a risk prediction method and system for construction. BACKGROUND

[0002] In the field of construction safety management, traditional methods mainly rely on manual inspection and static monitoring systems, which are difficult to capture the dynamic risk changes of the construction site in real time. The existing technology usually uses a single type of sensor to collect isolated data, lacking the ability of spatio-temporal collaborative analysis of multi-source heterogeneous data. At the same time, the experience-driven safety warning mechanism is difficult to effectively quantify the coupling risks of complex factors such as human-computer interaction, environmental mutation and structural deformation, resulting in lagging risk response and insufficient pertinence.

[0003] In recent years, although intelligent construction technology has gradually introduced building information model (BIM) and Internet of Things devices, there are still significant limitations in risk prediction. On the one hand, the discrete monitoring data is not deeply associated with the construction progress and spatial topology, and it is difficult to build a dynamic evolution risk map. On the other hand, conventional algorithms are difficult to handle high-dimensional spatio-temporal data such as vibration signals and point cloud models, and the prediction accuracy of complex accidents such as collapse and mechanical injury is insufficient. In addition, the existing warning strategy library lacks a self-adaptive adjustment mechanism and cannot dynamically optimize the decision threshold according to the characteristics of the construction stage.

[0004] In view of the above defects, it is urgent to build an intelligent analysis framework integrating multi-source dynamic data. The ideal solution needs to achieve three core breakthroughs: first, integrate structural monitoring, environmental parameters, personnel positioning and three-dimensional real scene data through spatio-temporal registration technology to form a unified analysis base; second, use deep learning models to mine the internal correlation of data and quantify the multi-dimensional risk coupling probability; third, establish a closed-loop optimization mechanism to dynamically adjust the prediction model and strategy library according to actual accident feedback, forming a continuously evolving active protection system. SUMMARY

[0005] According to a first aspect of the present application, the present application claims a risk prediction method for construction, comprising:

[0006] S1, acquiring multi-source dynamic data of a construction site to be predicted, and performing spatio-temporal fusion processing on the multi-source dynamic data to obtain a comprehensive data space of the construction site to be predicted;

[0007] S2, calculating a dynamic feature vector of the construction site to be predicted based on the comprehensive data space;

[0008] S3, inputting the dynamic feature vector into a deep neural network to output a risk probability distribution of the construction site to be predicted;

[0009] S4, matching the safety policy library of the current construction stage based on the risk probability distribution, generating a hierarchical early warning signal and a corresponding avoidance operation instruction set;

[0010] S5, dynamically adjusting the parameters of the deep neural network and updating the decision threshold of the safety policy library based on the difference between the actual risk event and the prediction result.

[0011] Further, the S1 further comprises:

[0012] High-precision vibration sensors are deployed on tower crane jibs, scaffolding nodes, and deep foundation support structures to capture real-time structural micro-deformation signals;

[0013] Fixed environmental monitoring stations continuously collect temperature, humidity, wind speed, and precipitation monitoring data;

[0014] A laser radar carried by a drone scans the construction site three times a week to generate a millimeter-level resolution three-dimensional point cloud model;

[0015] The component design size, material safety threshold, and critical path progress plan of the current construction stage are dynamically extracted from the building information modeling system;

[0016] Real-time location heat maps are generated by ultra-wideband positioning tags worn by construction personnel to distinguish different activity areas;

[0017] The origin of the building information model is used as the reference coordinate system, and a feature point matching algorithm is used to register the drone point cloud data to the design coordinate system;

[0018] A sequence of construction progress timestamps is added to the vibration sensor data stream to associate vibration events with construction processes;

[0019] Personnel positioning data is mapped to the spatial grid system of the building information model to establish a three-level location coding of "floor-work area-safety grid";

[0020] Four-dimensional data space is constructed by integrating environmental data and spatial coordinates, where the time dimension is associated with the meteorological change cycle and the construction shift.

[0021] Further, the S2 further comprises:

[0022] Calculate the structural deformation deviation, compare the coordinate offset of real-time point cloud and design model at key nodes, and calculate the cumulative deformation risk by combining the stress change trend monitored by intelligent bolt sensors;

[0023] Obtain the environmental fluctuation index, analyze the mutation amplitude and duration of temperature and humidity data based on a sliding time window, and identify persistent abnormal weather patterns;

[0024] Determine the intensity of human-computer interaction conflict by calculating the collision probability based on the density of the overlapping area between the tower crane hoisting trajectory envelope and the personnel positioning heat map, combined with the equipment tonnage parameters.

[0025] Calculate the risk value of schedule lag, identify the actual progress deviation of key processes such as concrete pouring, and correlate the current tower crane utilization rate with the status of labor allocation.

[0026] Furthermore, S3 also includes:

[0027] A 3D convolutional neural network is used to scan point cloud voxel data to identify spatial geometric risks;

[0028] By analyzing the periodic peak characteristics of wind speed sensor data through a gated loop unit, early warning of sudden strong wind accidents can be provided.

[0029] A dynamic graph neural network is constructed, where nodes include the type of construction workers, the working radius of the equipment, and the load-bearing capacity of the components, and the edge weights are updated in real time according to the number of people within the rotation range of the equipment.

[0030] Output a risk probability distribution map of the construction site layout, marking the level and impact range of three types of risks: collapse, fall from height, and mechanical injury.

[0031] Furthermore, S4 also includes:

[0032] The safety strategy library is invoked according to the type of construction stage. During the concrete curing period, the humidity threshold is the focus, and during the steel structure hoisting stage, the wind speed limit is strictly controlled.

[0033] A three-level early warning mechanism is generated: the first level triggers the audible and visual alarm, the second level freezes the tower crane operation permissions, and the third level activates the evacuation broadcast system.

[0034] The avoidance operation instruction set includes dynamic avoidance of areas with excessive structural stress by tower cranes, integration of personnel evacuation routes with real-time status of fire lanes, and automatic triggering of sprinkler dust suppression systems to deal with excessive dust.

[0035] Furthermore, S5 also includes:

[0036] When an actual safety accident occurs, the back-tracking prediction results generate a false alarm or missed alarm analysis report, highlighting the blind spots in mechanical injury identification.

[0037] By adjusting the weight allocation of neural network modules through reinforcement learning mechanisms, the identification priority of high-risk events such as collapses can be improved.

[0038] The threshold for the environmental fluctuation index is updated based on seasonal patterns, with the precipitation impact coefficient increased during the rainy season and the risk assessment of low-temperature freeze-thaw cycles strengthened during winter.

[0039] Furthermore, the method also includes:

[0040] The vibration sensor uses a triaxial accelerometer to monitor structural resonant frequency anomalies;

[0041] The environmental monitoring station integrates a rainfall detection unit to collect rainfall intensity data in real time;

[0042] Earth pressure sensors are deployed at key stress points of deep foundation pit support piles to monitor sudden changes in lateral pressure in the soil layer.

[0043] Spatial registration is performed by extracting feature points from the steel beam endpoints and column corners for position calibration.

[0044] The construction zoning coding system is divided into three levels: floor number, work surface function, and safety grid coordinates.

[0045] The timestamp synchronization uses a high-precision clock protocol to ensure millisecond-level alignment between sensor data and construction progress instructions.

[0046] Furthermore, the method also includes:

[0047] The elements fused for calculating structural deformation deviation include: displacement vector of cantilever scaffold anchorage points extracted from real-time point cloud, prestress loss data monitored by smart bolt sensors, construction load change trend converted from concrete pump truck working pressure, and weighted generation of comprehensive deformation risk index.

[0048] Anomalies in beam-column node displacements are identified by scanning the voxelized mesh of the point cloud using a 3D convolutional kernel.

[0049] The system detects persistent high-temperature patterns in environmental data and dynamically updates the spatial conflict probability between the tower crane hoisting trajectory envelope and personnel movement paths.

[0050] Furthermore, the method also includes:

[0051] The avoidance operation instruction set includes the coordinates of equipment shutdown and avoidance paths in high-risk areas, optimized navigation routes for personnel evacuation, and automatic start-up instruction sequences for protective facilities;

[0052] The avoidance operation instruction set includes tower crane avoidance path planning to avoid areas with excessive structural stress and densely populated areas, and personnel evacuation routes are integrated with real-time updates on the occupancy status of fire lanes;

[0053] The automatic start-up command sequence for protective facilities includes the activation of the automatic lifting baffle and the triggering of the safety net tension adjustment device; the node attributes of the dynamic graph neural network include the type of construction workers and their safety training level, the working radius and rated load parameters of the equipment, the design load-bearing level of the components and their current aging status; the edge weights are dynamically calculated based on the personnel density within the equipment's rotation radius and the duration of high-risk operations.

[0054] It adopts a gated loop unit structure, and the input window length is adaptively adjusted based on the construction stage type and environmental change detection results.

[0055] A confusion matrix between predicted results and actual accidents is established. The weight allocation of the neural network loss function is adjusted through a reinforcement learning mechanism, and the safety strategy library is periodically updated according to the evolution of the construction phase.

[0056] According to a second aspect of the present invention, the present invention claims protection for a risk prediction system for building construction, comprising:

[0057] One or more processors;

[0058] A memory storing one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the aforementioned risk prediction method for building construction.

[0059] This invention relates to the field of construction safety, specifically to a risk prediction method and system for construction projects. Addressing the shortcomings of existing technologies, such as isolated analysis of multi-source data, lag in dynamic risk response, and insufficient prediction accuracy, this invention constructs a spatiotemporal fusion data space, integrating multi-dimensional dynamic data including structural micro-deformation monitoring, environmental parameters, 3D real-scene scanning, personnel positioning, and building information models to form a unified analysis foundation. Based on a deep neural network architecture, a multimodal feature extraction mechanism is designed to quantify coupled risks and generate a partitioned risk probability distribution map. A safety strategy library is matched with the characteristics of each construction stage, implementing a three-level early warning mechanism and automated avoidance instructions. A closed-loop optimization mechanism is introduced, dynamically adjusting model parameters and decision thresholds based on feedback from actual accidents to achieve continuous evolution of the prediction system. This method significantly improves the proactive prevention and control capabilities for complex accidents such as collapses and falls from heights, enabling the construction of an adaptive intelligent protection system for construction sites. Attached Figure Description

[0060] Figure 1 A flowchart illustrating the workflow of a risk prediction method for building construction claimed in this application.

[0061] Figure 2 A second flowchart of a risk prediction method for building construction claimed in this application embodiment;

[0062] Figure 3 A third flowchart illustrating a risk prediction method for building construction claimed in this application.

[0063] Figure 4 The fourth flowchart of a risk prediction method for building construction claimed in this application is shown in an embodiment of the present application.

[0064] Figure 5 This is a structural diagram of a risk prediction system for building construction, which is claimed in an embodiment of this application. Detailed Implementation

[0065] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0066] The terms "first," "second," and "third" in this application are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first," "second," or "third" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of this application are only used to explain the relative positional relationships and movements between components in a specific orientation (as shown in the figures). If the specific orientation changes, the directional indications also change accordingly. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0067] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0068] According to a first embodiment of the present invention, the present invention claims protection for a risk prediction method for building construction, referring to... Figure 1 ,include:

[0069] S1, acquire multi-source dynamic data of the construction site with the risk to be predicted, and perform spatiotemporal fusion processing on the multi-source dynamic data to obtain the comprehensive data space of the construction site with the risk to be predicted;

[0070] S2, based on the comprehensive data space, calculate the dynamic feature vector of the construction site with the risk to be predicted;

[0071] S3, input the dynamic feature vector into the deep neural network, and output the risk probability distribution of the construction site to be predicted;

[0072] S4. Based on the risk probability distribution, match the safety strategy library of the current construction stage to generate graded early warning signals and corresponding avoidance operation instruction sets.

[0073] S5, Based on the difference between the actual risk event and the prediction result, dynamically adjust the parameters of the deep neural network and update the decision threshold of the security policy library.

[0074] This method was implemented in the construction of a 380-meter super high-rise project. The project included a deep foundation pit (22 meters deep), a large luffing jib tower crane (50-ton load capacity), and high-altitude cantilever operations (maximum cantilever length 8 meters). The implementation period was from March 2024 to January 2025, covering the rainy season, typhoon season, and the critical stage of the main structure's topping-out.

[0075] Furthermore, S1 also includes:

[0076] High-precision vibration sensors are deployed on tower crane booms, scaffolding nodes, and deep foundation pit support structures to capture micro-deformation signals of the structures in real time.

[0077] Temperature, humidity, wind speed and precipitation monitoring data are continuously collected through fixed environmental monitoring stations;

[0078] Using drones equipped with lidar, the construction site is scanned three times a week to generate a 3D point cloud model with millimeter-level resolution;

[0079] Dynamically extract component design dimensions, material safety thresholds, and critical path schedules for the current construction phase from the building information modeling system;

[0080] Real-time location heatmaps are generated by using ultra-wideband positioning tags worn by construction workers to distinguish the activity areas of different types of work.

[0081] Using the origin of the building information model as the reference coordinate system, a feature point matching algorithm is used to register the UAV point cloud data to the design coordinate system;

[0082] Add a construction progress timestamp sequence to the vibration sensor data stream to associate vibration events with construction procedures;

[0083] Personnel location data is mapped to the spatial grid system of the building information model to establish a three-level location code of "floor-work area-safety grid";

[0084] A four-dimensional data space is constructed by integrating environmental data and spatial coordinates, with the time dimension being associated with meteorological change cycles and construction shifts.

[0085] In this embodiment, multi-source dynamic data acquisition is implemented using IoT sensing deployment. 120 triaxial vibration sensors with a sensitivity of ±0.01g are arranged in the core tube shear wall to monitor the resonance caused by concrete pouring vibration. 40 earth pressure gauges with a range of 0-50MPa are installed on the foundation pit support piles to record changes in soil lateral pressure during the typhoon season. The environmental monitoring station collects temperature, humidity, and wind speed every 5 minutes, with a maximum wind speed of 35m / s recorded during the passage of Typhoon Talim.

[0086] Spatial data acquisition includes point cloud generation by UAV scanning three times a week with an accuracy of ±3mm, and a total of 82 point cloud model versions generated; the BIM system updates design parameters in real time, such as design changes that increased the thickness of the core tube steel plate from 30mm to 35mm.

[0087] During the personnel positioning, 412 workers wore UWB tags with a positioning accuracy of ±15cm. The generated heat map showed that the density of steelworkers on the L52 floor reached 0.8 people / ㎡.

[0088] Furthermore, referring to Figure 2 S2 further includes:

[0089] Calculate the structural deformation deviation, compare the coordinate offset of the real-time point cloud with the design model at key nodes, and calculate the cumulative deformation risk by combining the stress change trend monitored by the smart bolt sensor.

[0090] Obtain environmental fluctuation indices, analyze the magnitude and duration of abrupt changes in temperature and humidity data based on sliding time windows, and identify persistent abnormal weather patterns;

[0091] Determine the intensity of human-computer interaction conflict by calculating the collision probability based on the density of the overlapping area between the tower crane hoisting trajectory envelope and the personnel positioning heat map, combined with the equipment tonnage parameters.

[0092] Calculate the risk value of schedule lag, identify the actual progress deviation of key processes such as concrete pouring, and correlate the current tower crane utilization rate with the status of labor allocation.

[0093] In this embodiment, 28 core cylinder corner points were selected as feature points during point cloud registration, and the registration error was controlled within ±5mm. During the typhoon, the maximum deviation between the point cloud and the BIM model was recorded as 17mm, located at the northwest corner cantilever.

[0094] In the data mapping, layer L52 was divided into 36 4m×4m safety grids. The vibration data timestamps were synchronized with the concrete pouring records, and 23 vibration timeout events were found.

[0095] Furthermore, referring toFigure 3 S3 further includes:

[0096] A 3D convolutional neural network is used to scan point cloud voxel data to identify spatial geometric risks;

[0097] By analyzing the periodic peak characteristics of wind speed sensor data through a gated loop unit, early warning of sudden strong wind accidents can be provided.

[0098] A dynamic graph neural network is constructed, where nodes include the type of construction workers, the working radius of the equipment, and the load-bearing capacity of the components, and the edge weights are updated in real time according to the number of people within the rotation range of the equipment.

[0099] Output a risk probability distribution map of the construction site layout, marking the level and impact range of three types of risks: collapse, fall from height, and mechanical injury.

[0100] In this embodiment, the structural deformation characteristics are as follows: the maximum displacement monitored at the anchorage point of the L60 layer cantilever frame is 9.8 mm, with a design threshold of 10 mm; the smart bolt sensor shows a prestress loss of 12%, with a warning threshold of 15%.

[0101] The environmental fluctuation index indicates that the standard deviation of temperature and humidity during the continuous high-temperature period reached 2.3 times the historical average.

[0102] The intensity of human-machine conflict was 18%, with the peak overlap rate between the tower crane hoisting path and the steel reinforcement worker's heat map occurring during the installation stage of the steel plate wall on floor L48.

[0103] Furthermore, referring to Figure 4 S4 further includes:

[0104] The safety strategy library is invoked according to the type of construction stage. During the concrete curing period, the humidity threshold is the focus, and during the steel structure hoisting stage, the wind speed limit is strictly controlled.

[0105] A three-level early warning mechanism is generated: the first level triggers the audible and visual alarm, the second level freezes the tower crane operation permissions, and the third level activates the evacuation broadcast system.

[0106] The avoidance operation instruction set includes dynamic avoidance of areas with excessive structural stress by tower cranes, integration of personnel evacuation routes with real-time status of fire lanes, and automatic triggering of sprinkler dust suppression systems to deal with excessive dust.

[0107] Furthermore, S5 also includes:

[0108] When an actual safety accident occurs, the back-tracking prediction results generate a false alarm or missed alarm analysis report, highlighting the blind spots in mechanical injury identification.

[0109] By adjusting the weight allocation of neural network modules through reinforcement learning mechanisms, the identification priority of high-risk events such as collapses can be improved.

[0110] The threshold for the environmental fluctuation index is updated based on seasonal patterns, with the precipitation impact coefficient increased during the rainy season and the risk assessment of low-temperature freeze-thaw cycles strengthened during winter.

[0111] In this embodiment, a three-level early warning trigger is adopted. On September 15, 2024, the environmental module detected a sudden increase in wind speed to 28 m / s, exceeding the second-level threshold of 25 m / s. The system then froze the operation permissions of tower crane #3 and initiated the AR projection of a virtual warning line.

[0112] When optimizing evacuation routes, the traditional route, from the east side of floor L50 to the core tube staircase, takes 8 minutes; while the optimized route adds an emergency passage through a cantilevered scaffold, shortening the time to 5 minutes.

[0113] Furthermore, the method also includes:

[0114] The vibration sensor uses a triaxial accelerometer to monitor structural resonant frequency anomalies;

[0115] The environmental monitoring station integrates a rainfall detection unit to collect rainfall intensity data in real time;

[0116] Earth pressure sensors are deployed at key stress points of deep foundation pit support piles to monitor sudden changes in lateral pressure in the soil layer.

[0117] Spatial registration is performed by extracting feature points from the steel beam endpoints and column corners for position calibration.

[0118] The construction zoning coding system is divided into three levels: floor number, work surface function, and safety grid coordinates.

[0119] The timestamp synchronization uses a high-precision clock protocol to ensure millisecond-level alignment between sensor data and construction progress instructions.

[0120] Furthermore, the method also includes:

[0121] The elements fused for calculating structural deformation deviation include: displacement vector of cantilever scaffold anchorage points extracted from real-time point cloud, prestress loss data monitored by smart bolt sensors, construction load change trend converted from concrete pump truck working pressure, and weighted generation of comprehensive deformation risk index.

[0122] Anomalies in beam-column node displacements are identified by scanning the voxelized mesh of the point cloud using a 3D convolutional kernel.

[0123] The system detects persistent high-temperature patterns in environmental data and dynamically updates the spatial conflict probability between the tower crane hoisting trajectory envelope and personnel movement paths.

[0124] In this embodiment, 1,285 sets of data from the Hong Kong International Airport accident database were used for model training and execution pre-training. The spatial topology module identified that the tilt angle of the L55 layer support frame exceeded the standard.

[0125] For the dynamic calculation of the graph neural network, the working radius of the tower crane node is 45m, the lifting weight is 32 tons, the density of steelworkers on the west side of the L50 floor is 0.6 people / ㎡, the collision risk probability output is 24.7%, which exceeds the threshold by 20%.

[0126] When the window length is adaptive, a 24-hour window length is used for concrete strength growth analysis during the main construction phase; during typhoon warnings, the window length is automatically switched to 5 minutes to monitor sudden wind speed changes.

[0127] Furthermore, the method also includes:

[0128] The avoidance operation instruction set includes the coordinates of equipment shutdown and avoidance paths in high-risk areas, optimized navigation routes for personnel evacuation, and automatic start-up instruction sequences for protective facilities;

[0129] The avoidance operation instruction set includes tower crane avoidance path planning to avoid areas with excessive structural stress and densely populated areas, and personnel evacuation routes are integrated with real-time updates on the occupancy status of fire lanes;

[0130] The automatic start-up command sequence for protective facilities includes the activation of the automatic lifting baffle and the triggering of the safety net tension adjustment device; the node attributes of the dynamic graph neural network include the type of construction workers and their safety training level, the working radius and rated load parameters of the equipment, the design load-bearing level of the components and their current aging status; the edge weights are dynamically calculated based on the personnel density within the equipment's rotation radius and the duration of high-risk operations.

[0131] It adopts a gated loop unit structure, and the input window length is adaptively adjusted based on the construction stage type and environmental change detection results.

[0132] A confusion matrix between predicted results and actual accidents is established. The weight allocation of the neural network loss function is adjusted through a reinforcement learning mechanism, and the safety strategy library is periodically updated according to the evolution of the construction phase.

[0133] In this embodiment, the input window length adaptive mechanism of the gated loop unit is as follows: during the main structure construction stage, a 24-hour long cycle mode is used to analyze the concrete strength growth trend; during the high-altitude operation stage, the mode is switched to a 5-minute short cycle mode to monitor sudden gusts; and when a sudden change in environmental data is detected, a 10-second high-speed sampling window is automatically activated.

[0134] The closed-loop optimization uses a confusion matrix to distinguish between accident types such as false alarms of mechanical injuries and missed alarms of collapses; a reinforcement learning reward function prioritizes optimizing the recognition rate of high-risk events; the safety strategy library is updated quarterly based on construction characteristics, with enhanced pit monitoring thresholds during the rainy season and the addition of anti-slip measures decision items in winter. The solution also includes a visual early warning system that maps the risk probability distribution to a 3D construction model, using a red-yellow-blue gradient to indicate risk levels; augmented reality equipment projects virtual warning lines and evacuation direction arrows to on-site personnel; and a construction command platform generates dynamic heatmaps to display risk evolution trends and simulates animations of different response plans.

[0135] Visualized early warning maps the risk probability distribution onto a 3D construction model, using a red-yellow-blue gradient to indicate risk levels; augmented reality equipment projects virtual warning lines and evacuation direction arrows to on-site personnel; the construction command platform generates dynamic heat maps to display risk evolution trends and simulates the effects of different response plans with animation.

[0136] According to a second embodiment of the present invention, the present invention claims protection for a risk prediction system for building construction, referring to... Figure 5 ,include:

[0137] One or more processors;

[0138] A memory storing one or more programs, which, when executed by one or more processors, enable the one or more processors to implement the aforementioned risk prediction method for building construction.

[0139] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0140] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0141] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A risk prediction method for building construction, characterized in that, include: S1, acquire multi-source dynamic data of the construction site with the risk to be predicted, and perform spatiotemporal fusion processing on the multi-source dynamic data to obtain the comprehensive data space of the construction site with the risk to be predicted; S2, based on the comprehensive data space, calculate the dynamic feature vector of the construction site with the risk to be predicted; S3, input the dynamic feature vector into the deep neural network, and output the risk probability distribution of the construction site to be predicted; S4. Based on the risk probability distribution, match the safety strategy library of the current construction stage to generate graded early warning signals and corresponding avoidance operation instruction sets. S5, Based on the difference between the actual risk event and the prediction result, dynamically adjust the parameters of the deep neural network and update the decision threshold of the security policy library.

2. The risk prediction method for building construction according to claim 1, characterized in that, S1 further includes: High-precision vibration sensors are deployed on tower crane booms, scaffolding nodes, and deep foundation pit support structures to capture micro-deformation signals of the structures in real time. Temperature, humidity, wind speed and precipitation monitoring data are continuously collected through fixed environmental monitoring stations; the construction site is scanned three times a week using drones equipped with lidar to generate a three-dimensional point cloud model with millimeter-level resolution. Dynamically extract component design dimensions, material safety thresholds, and critical path schedules for the current construction phase from the building information modeling system; Real-time location heatmaps are generated by using ultra-wideband positioning tags worn by construction workers to distinguish the activity areas of different types of work. Using the origin of the building information model as the reference coordinate system, a feature point matching algorithm is used to register the UAV point cloud data to the design coordinate system; Add a construction progress timestamp sequence to the vibration sensor data stream to associate vibration events with construction procedures; Personnel location data is mapped to the spatial grid system of the building information model to establish a three-level location code of "floor-work area-safety grid"; A four-dimensional data space is constructed by integrating environmental data and spatial coordinates, with the time dimension being associated with meteorological change cycles and construction shifts.

3. The risk prediction method for building construction according to claim 1, characterized in that, The S2 further includes: Calculate the structural deformation deviation, compare the coordinate offset of the real-time point cloud and the design model at key nodes, and calculate the cumulative deformation risk by combining the stress change trend monitored by the smart bolt sensor; obtain the environmental fluctuation index, analyze the abrupt change amplitude and duration of temperature and humidity data based on the sliding time window, and identify persistent abnormal weather patterns. Determine the intensity of human-computer interaction conflict by calculating the collision probability based on the density of the overlapping area between the tower crane hoisting trajectory envelope and the personnel positioning heat map, combined with the equipment tonnage parameters. Calculate the risk value of schedule lag, identify the actual progress deviation of key processes such as concrete pouring, and correlate the current tower crane utilization rate with the status of labor allocation.

4. The risk prediction method for building construction according to claim 1, characterized in that, The S3 further includes: A 3D convolutional neural network is used to scan point cloud voxel data to identify spatial geometric risks; By analyzing the periodic peak characteristics of wind speed sensor data through a gated loop unit, early warning of sudden strong wind accidents can be provided. A dynamic graph neural network is constructed, where nodes include the type of construction workers, the working radius of the equipment, and the load-bearing capacity of the components, and the edge weights are updated in real time according to the number of people within the rotation range of the equipment. Output a risk probability distribution map of the construction site layout, marking the level and impact range of three types of risks: collapse, fall from height, and mechanical injury.

5. The risk prediction method for building construction according to claim 1, characterized in that, The S4 further includes: The safety strategy library is invoked according to the type of construction stage. During the concrete curing period, the humidity threshold is the focus, and during the steel structure hoisting stage, the wind speed limit is strictly controlled. A three-level early warning mechanism is generated: the first level triggers the audible and visual alarm, the second level freezes the tower crane operation permissions, and the third level activates the evacuation broadcast system. The avoidance operation instruction set includes dynamic avoidance of areas with excessive structural stress by tower cranes, integration of personnel evacuation routes with real-time status of fire lanes, and automatic triggering of sprinkler dust suppression systems to deal with excessive dust.

6. The risk prediction method for building construction according to claim 1, characterized in that, The S5 also includes: When an actual safety accident occurs, the back-tracking prediction results generate a false alarm or missed alarm analysis report, highlighting the blind spots in mechanical injury identification. By adjusting the weight allocation of neural network modules through reinforcement learning mechanisms, the identification priority of high-risk events such as collapses can be improved. The threshold for the environmental fluctuation index is updated based on seasonal patterns, with the precipitation impact coefficient increased during the rainy season and the risk assessment of low-temperature freeze-thaw cycles strengthened during winter.

7. A risk prediction method for building construction according to claim 2, characterized in that, Also includes: The vibration sensor uses a triaxial accelerometer to monitor structural resonant frequency anomalies; The environmental monitoring station integrates a rainfall detection unit to collect rainfall intensity data in real time; Earth pressure sensors are deployed at key stress points of deep foundation pit support piles to monitor sudden changes in lateral pressure in the soil layer. Spatial registration is performed by extracting feature points from the steel beam endpoints and column corners for position calibration. The construction zoning coding system is divided into three levels: floor number, work surface function, and safety grid coordinates; the timestamp synchronization adopts a high-precision clock protocol to ensure millisecond-level alignment between sensor data and construction progress instructions.

8. A risk prediction method for building construction according to claim 2, characterized in that, Also includes: The elements fused for calculating structural deformation deviation include: displacement vector of cantilever scaffold anchorage points extracted from real-time point cloud, prestress loss data monitored by smart bolt sensors, construction load change trend converted from concrete pump truck working pressure, and weighted generation of comprehensive deformation risk index. Anomalies in beam-column node displacements are identified by scanning the voxelized mesh of the point cloud using a 3D convolutional kernel. The system detects persistent high-temperature patterns in environmental data and dynamically updates the spatial conflict probability between the tower crane hoisting trajectory envelope and personnel movement paths.

9. A risk prediction method for building construction according to claim 2, characterized in that, Also includes: The avoidance operation instruction set includes equipment shutdown coordinates and avoidance paths in high-risk areas, optimized navigation routes for personnel evacuation, and automatic start-up instruction sequences for protective facilities; the avoidance operation instruction set includes tower crane avoidance path planning to avoid areas with excessive structural stress and densely populated areas, and personnel evacuation routes integrating real-time updates on the occupancy status of fire lanes; The automatic start-up command sequence for protective facilities includes the activation of the automatic lifting baffle and the triggering of the safety net tension adjustment device; the node attributes of the dynamic graph neural network include the type of construction workers and their safety training level, the working radius and rated load parameters of the equipment, the design load-bearing level of the components and their current aging status; the edge weights are dynamically calculated based on the personnel density within the equipment's rotation radius and the duration of high-risk operations. It adopts a gated loop unit structure, and the input window length is adaptively adjusted based on the construction stage type and environmental change detection results. A confusion matrix between predicted results and actual accidents is established. The weight allocation of the neural network loss function is adjusted through a reinforcement learning mechanism, and the safety strategy library is periodically updated according to the evolution of the construction phase.

10. A risk prediction system for building construction, characterized in that, include: One or more processors; A memory having stored one or more programs that, when executed by one or more processors, cause the one or more processors to implement a risk prediction method for building construction according to any one of claims 1 to 9.

Images