Construction safety intelligent management system and method based on artificial intelligence

Abstract

The invention discloses a construction safety intelligent management system and method based on artificial intelligence, and relates to the technical field of construction safety, and the method comprises the steps: collecting body motion data, spatial positioning data and environment interaction data of an operator through an intelligent terminal, and constructing a multi-source data set; the system generates a multi-scale feature set through extraction of micro-action features, macroscopic behavior features and environment interaction features, further calculates an unbalance falling risk, a high falling risk and a collision injury risk of an operator, and comprehensively evaluates an instantaneous risk value; calculating a total risk value in combination with the risk attenuation function and the environmental static risk value, carrying out graded early warning according to the total risk value and the change trend thereof, and notifying an operator through the intelligent terminal; the invention aims to monitor the safety risk of operating personnel in real time, give out early warning in time, effectively reduce the safety accident risk in building operation and improve the safety of the operating personnel.

Description

Technical Field

[0001] This invention belongs to the field of construction safety technology, specifically, it relates to an intelligent management system and method for construction safety based on artificial intelligence. Background Technology

[0002] Construction site environments are complex and changeable, and workers face many safety risks, such as loss of balance and falls, falls from heights, and collision injuries.

[0003] Existing construction safety technologies typically suffer from significant drawbacks, including low integration, limited perception dimensions, static risk assessments, and rudimentary early warning mechanisms. Current systems rely on single data sources, such as using location data to monitor personnel entering hazardous areas or relying solely on inertial sensors to detect falls. This fragmented approach fails to comprehensively capture the complex and dynamic risk coupling at the work site, and cannot effectively identify the composite risks arising from subtle postural instability, abnormal behavioral patterns, and dynamic interactions with environmental obstacles. Furthermore, existing risk assessment models are often static and lagging, relying primarily on pre-set fixed rules or simple thresholds for judgment, such as only assessing risks when personnel cross... When electronic fences trigger alarms, they lack the ability to quantitatively model and dynamically calculate the continuous changes in personnel's physical state, the spatiotemporal patterns of their behavioral trajectories, and the potential impact of historical risk locations on their current state. This results in insufficient forward-looking prediction and identification of gradual risk evolution. Often, the system only triggers a response when the risk has already significantly increased or an accident is about to occur, thus missing the golden window for early warning and intervention. In addition, existing early warning methods are usually relatively simple and general, failing to provide refined and differentiated tiered push notifications based on the severity, urgency, and changing trends of the risk. This can easily lead to alarm fatigue among workers due to frequent false alarms or alarms of a single intensity, causing them to ignore real dangers.

[0004] To address the aforementioned problems, this invention proposes an intelligent management system and method for construction safety based on artificial intelligence. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent management system and method for construction safety based on artificial intelligence, which solves the problems of incomplete dynamic and complex risk perception, delayed assessment, and crude early warning in the field of construction safety technology.

[0006] The objective of this invention can be achieved through the following technical solutions: An artificial intelligence-based intelligent management method for construction safety, the method comprising: Step 1: Based on the smart terminal worn by the workers, collect the workers' body motion data, spatial positioning data and environmental interaction data, and perform spatiotemporal alignment to construct a multi-source data set; Step 2: Based on the constructed multi-source dataset, simultaneously perform micro-action feature extraction, macro-behavioral feature extraction, and environmental interaction feature extraction to generate a multi-scale feature set; Step 3: Based on the multi-scale feature set, calculate the current associated components of the worker's risk of imbalance and fall, risk of fall from height, and risk of collision injury, and comprehensively assess the worker's instantaneous comprehensive risk value. Step 4: Construct a risk attenuation function that is tied to the workers, assess the interactive risk value of the workers in three-dimensional space, and calculate the total risk value of the workers by combining the static environmental risk value and the instantaneous comprehensive risk value. Step 5: Based on the total risk value and its changing trend, execute the tiered early warning operation and notify the operators of the early warning through the smart terminal.

[0007] As a further embodiment of the present invention, in step one, the motion data of the body is collected based on the nine-axis inertial measurement unit integrated in the smart terminal, including three-axis acceleration ACC(t), three-axis angular velocity GYRO(t) and three-axis magnetic field strength MAG(t), where (t) represents the time-related data. Spatial positioning data is collected based on the UWB ultra-wideband positioning unit and inertial navigation module integrated in the smart terminal, including three-dimensional coordinates [X(t), Y(t), Z(t)]. The environmental interaction data is collected based on the millimeter-wave radar integrated into the smart terminal, including the relative distance D_obj(t), relative velocity V_obj(t), and azimuth angle θ_obj(t) of obstacles within a 5-meter distance centered on the operator; Body motion data, spatial positioning data, and environmental interaction data are collected based on a preset uniform sampling time interval.

[0008] As a further aspect of the present invention, the specific method for constructing the multi-source data set in step one is as follows: Time alignment is performed based on the time points bound to the body motion data, spatial positioning data, and environmental interaction data, respectively. The time-aligned body motion data, spatial positioning data, and environmental interaction data are converted into a pre-built 3D panoramic model of the construction site and mapped to the body coordinate system with the centroid of the worker as the origin. The body motion data, spatial positioning data, and environmental interaction data in the body coordinate system are extracted and combined into a multi-source data set.

[0009] As a further aspect of the present invention, the specific method for generating the multi-scale feature set in step two is as follows: S1, Micro-motion feature extraction: S11, calculate the standard deviation σ_A and kurtosis Kurt_A of the acceleration vector sum A_sum=sqrt(ACC_x²+ACC_y²+ACC_z²) within the time window T1, where sqrt() is the square root function, and the time window T1 is the time period preset by the operator; S12, calculate the real-time pitch angle α_pitch(t) and roll angle α_roll(t) of the worker's torso relative to the ground based on the quaternion method; S13, calculate the attitude stability index S_posture=1 / (1+σ_α_pitch+σ_α_roll), where σ_α is the standard deviation of the angle within 1 second, σ_α_pitch represents the standard deviation of the real-time pitch angle within 1 second, and σ_α_roll represents the standard deviation of the roll angle within 1 second; S2, Macro-behavioral feature extraction: S21. Based on the three-dimensional coordinates [X(t),Y(t),Z(t)] of the operator, determine the movement trajectory of the operator within the time window T1, and calculate the rate of change of curvature of the movement trajectory: C_curve(t)=|Δθ_heading / ΔS|, where Δθ_heading is the change in heading angle and ΔS is the movement arc length; S22, calculate the motion entropy H_motion of the worker in three-dimensional space = -Σ(p_i*log2(p_i)), where p_i is the probability of the velocity vector in each discrete direction interval on the unit sphere; S3, Environmental Interaction Feature Extraction: S31, when the millimeter-wave radar detects that D_obj(t) < the preset safe distance D_s, the following is adopted: γ_approach=(D_s-D_obj(t)) / D_s+(V_obj(t) / V_s); Calculate the approach hazard factor γ_approach, where V_s is the preset reference speed; Extract micro-motion features, including acceleration vector and A_sum, standard deviation σ_A, kurtosis Kurt_A, real-time pitch angle α_pitch(t), roll angle α_roll(t), and attitude stability index S_posture; Extract macroscopic behavioral features, including the rate of change of curvature C_curve(t) and the motion entropy H_motion; Extract environmental interaction features, including the proximity hazard factor γ_approach; The micro-action features, macro-behavioral features, and environmental interaction features are combined to generate a multi-scale feature set F(t).

[0010] As a further aspect of the present invention, the specific method for comprehensively evaluating the instantaneous comprehensive risk value of the operators in step three is as follows: S51, Calculate the imbalance fall risk component R_stumble(t): R_stumble(t)=w1*(1-S_posture(t))+w2*(σ_A(t) / σ_A_max)+w3*(|Kurt_A(t)-3| / Kurt_norm); Where w1, w2, w3 are preset weight coefficients, and w1+w2+w3=1, σ_A_max is a preset acceleration standard deviation threshold, and Kurt_norm is a kurtosis reference value; S52, Calculate the fall risk component R_fall(t): R_fall(t)=f_height(Z(t))*[λ1*g_dist(D_edge(t))+λ2*(H_motion(t) / H_max)+λ3*(C_curve(t) / C_max)]; Where f_height(Z(t)) is the height penalty function, f_height(Z(t))=1+log(1+Z(t) / Z_ref), and Z_ref is the preset reference height; D_edge(t) is the projected horizontal distance from the worker to the nearest danger edge, calculated in real time based on the worker's three-dimensional coordinates [X(t),Y(t),Z(t)] and a pre-built three-dimensional panoramic model of the construction site. g_dist(D_edge(t)) is the edge distance risk function, g_dist(D_edge(t))=exp(-D_edge(t) / D_safe), where D_safe is the preset safe distance threshold; λ1, λ2, λ3 are preset weighting coefficients, and λ1+λ2+λ3=1. H_max and C_max are preset motion entropy threshold and curvature change rate threshold, respectively. S53, Calculate the collision damage risk component R_collision(t): R_collision(t)=max(γ_approach(t))*[1+η*(||V_worker(t)|| / V_worker_max)]; Where max(γ_approach(t)) is the maximum value of the approach hazard coefficients of all obstacles calculated from millimeter-wave radar data at time t, ||V_worker(t)|| is the speed of the worker at time t calculated based on spatial positioning data, V_worker_max is the preset safe speed limit for the worker, and η is the preset adjustment coefficient. S54, based on the imbalance fall risk component R_stumble(t), the height fall risk component R_fall(t), and the collision injury risk component R_collision(t), calculate the instantaneous comprehensive risk value R_instant(t): R_instant(t)=β1*R_stumble(t)+β2*R_fall(t)+β3*R_collision(t); Wherein, β1, β2, β3 are preset comprehensive weight coefficients, and β1+β2+β3=1.

[0011] As a further aspect of the present invention, in step four, the specific method for constructing a risk attenuation function bound to the operator and assessing the interactive risk value of the operator in three-dimensional space is as follows: Create a personal historical risk location record list L_risk for each operator, where any record contains the three-dimensional coordinates of the risk location [X_risk, Y_risk, Z_risk]. The three-dimensional coordinates of the risk location correspond to the initial fall risk component R_fall_init and the record generation timestamp T_risk; When the instantaneous fall risk component R_fall(t) exceeds the preset activation threshold R_fall_th, the spatial positioning data [X(t),Y(t),Z(t)] of the worker at the current moment is used as the three-dimensional coordinates of the risk location, and R_fall(t) is used as the initial fall risk component R_fall_init. Together with the current timestamp, a new record is formed and added to L_risk. Construct the risk decay function Φ(Δt,R_init) that is bound to the operator, and calculate the risk value of any risk location after the three-dimensional coordinates decay over time: R_decayed=Φ(Δt,R_init)=R_init*exp(-k*Δt) Where Δt is the time difference between the current assessment time and the risk record timestamp T_risk, R_init is the initial high-altitude fall risk component R_fall_init, k is the preset attenuation coefficient, and exp() is the exponential function; Based on the spatial positioning data of the operator at the current moment [X(t),Y(t),Z(t)], calculate the distance d_i to all historical risk locations in L_risk = sqrt((X(t)-X_risk)²+(Y(t)-Y_risk)²+(Z(t)-Z_risk)²); Iterate through L_risk, and for each record whose distance from the current position d_i is less than the preset interaction radius R_int, calculate its decayed risk value R_decayed based on the risk decay function Φ(Δt,R_init); Define the maximum value among all risk values ​​R_decayed as the interaction risk value R_interaction(t) of the operator at the current moment: If there is no record that satisfies d_i < R_int, let R_interaction(t) = 0.

[0012] As a further aspect of the present invention, the specific method for calculating the total risk value of the workers in step four, combining the static environmental risk value and the instantaneous comprehensive risk value, is as follows: S71, based on the operator's current three-dimensional coordinates [X(t),Y(t),Z(t)], query the pre-constructed environmental static risk map of the operator and obtain the environmental static risk value R_environment(t) corresponding to the three-dimensional coordinates; Among them, the environmental static risk map is constructed based on the three-dimensional panoramic model of the construction site, and predefines the risk level values ​​related to static environmental hazards for different spatial areas in the model. S72, combining the instantaneous comprehensive risk value R_instant(t), the interaction risk value R_interaction(t), and the static environmental risk value R_environment(t), calculate the total risk value R_total(t) for the workers: R_total(t)=γ1*R_instant(t)+γ2*R_interaction(t)+γ3*R_environment(t) Wherein, γ1, γ2, and γ3 are preset fusion weight coefficients, and γ1+γ2+γ3=1.

[0013] As a further aspect of the present invention, the specific method for performing the tiered early warning operation based on the total risk value and its changing trend in step five is as follows: S81, extract the risk warning threshold, including the first threshold Th_low, the second threshold Th_medium and the third threshold Th_high, where 0 < Th_low < Th_medium < Th_high ≤ 1, and [0,1] represents the risk warning threshold range. S82, calculate the rate of change V_risk(t) of the total risk value R_total(t) within a preset time window T2, where the duration of T2 is greater than that of T1; S83, based on the threshold range to which the total risk value R_total(t) belongs and its rate of change V_risk(t), determine the warning level, including attention level, warning level, and critical level: If R_total(t)≥Th_high or R_total(t)≥Th_medium and V_risk(t)>preset rise rate threshold V_up_th, it is judged as a critical level warning; If Th_medium≤R_total(t)<Th_high and V_risk(t)≤V_up_th, or Th_low≤R_total(t)<Th_medium and V_risk(t)>V_up_th, it is determined to be a warning level alert; If Th_low≤R_total(t)<Th_medium and V_risk(t)≤V_up_th, it is determined to be a warning at the attention level; If R_total(t) < Th_low, then it is determined to be a state without warning; When S84 is determined to be a warning level, a low-frequency intermittent vibration and a yellow visual alert are emitted through the smart terminal. When the warning level is determined, a continuous vibration, a red visual alert, and a level one buzzer sound will be emitted through the smart terminal. When a critical level warning is issued, the system will send a high-intensity continuous vibration, a flashing red visual alert, and a level two sharp buzzer sound through the smart terminal, and simultaneously send alarm information including the three-dimensional coordinates of the operator and the warning level to the operator.

[0014] An artificial intelligence-based intelligent management system for construction safety, the system comprising: The multi-source spatiotemporal alignment acquisition module, based on the smart terminal worn by the operator, collects the operator's associated body motion data, spatial positioning data and environmental interaction data, and performs spatiotemporal alignment to construct a multi-source data set; The multi-scale feature fusion and extraction module, based on the constructed multi-source dataset, simultaneously performs micro-action feature extraction, macro-behavioral feature extraction, and environmental interaction feature extraction to generate a multi-scale feature set. The instantaneous risk component calculation module, based on a multi-scale feature set, calculates the current associated risk components of imbalance and fall, fall from height, and collision injury for workers, and comprehensively assesses the instantaneous comprehensive risk value of workers. The spatiotemporal interactive risk integration module constructs a risk attenuation function bound to the operator, assesses the interactive risk value of the operator in three-dimensional space, and calculates the total risk value of the operator by combining the static environmental risk value and the instantaneous comprehensive risk value. The tiered early warning push module executes tiered early warning operations based on the total risk value and its changing trend, and notifies the operators of the early warning notification through smart terminals.

[0015] The beneficial effects of this invention are: This invention integrates multi-source sensors to achieve multi-dimensional synchronous perception and data fusion of people, environment, and behavior, constructing a precise and real-time dynamic safety risk assessment system. Its advantages lie in realizing full-scale risk identification from microscopic actions to macroscopic behaviors and environmental interactions, enhancing the comprehensiveness and foresight of risk warnings. By fusing spatiotemporally aligned multi-source data and combining it with a personalized risk attenuation model, it can accurately quantify the dynamic comprehensive risks of workers in three-dimensional space, effectively identifying various hazards such as falls, falls from heights, and collisions. Finally, relying on a tiered early warning mechanism, it enables immediate risk intervention, thereby improving the initiative and accuracy of on-site safety management and effectively ensuring personnel safety. This invention achieves comprehensive, three-dimensional, and refined perception of workers' status by constructing a multi-scale feature set. Its advantage lies in integrating multi-dimensional information, from microscopic bodily control to macroscopic movement patterns and external environmental interactions, thereby improving the accuracy of behavioral analysis and safety risk assessment. Specifically, micro-motion features can keenly capture subtle postural tremors reflecting fatigue or distraction; macroscopic behavioral features quantify the orderliness and intent of movement at the spatial trajectory level; and environmental interaction features dynamically assess real-time external risks. This multi-source, multi-scale feature fusion overcomes the limitations of single data sources, making the judgment of personnel status more accurate and the early warning more timely. This invention achieves dynamic and accurate perception of the safety status of workers through multi-source data fusion and hierarchical calculation. Its advantages lie in integrating posture, position, motion, and environmental data, quantifying the three main risks of imbalance, fall from height, and collision into interrelated components, and enhancing the contextual relevance of risk perception through mechanisms such as height penalty and edge distance function. By adjusting the influence of various factors through weight coefficients, it achieves adaptive assessment of different work scenarios. The resulting instantaneous comprehensive risk value can comprehensively and in real time reflect the overall danger level of personnel, thereby improving the accuracy and timeliness of early warning. This invention improves the accuracy and foresight of safety risk assessment for high-altitude operations by constructing a dynamic and three-dimensional risk perception system. Its advantages lie in the innovative introduction of a spatiotemporal dimension, which not only captures the instantaneous behavioral risks of workers in real time but also records and quantifies the impact of historically high-risk locations through a risk decay function. This allows risk assessment to reflect the accumulation and dissipation process of risks, better reflecting the continuity of risks in real-world work scenarios. Simultaneously, it organically integrates three risk sources: personnel behavior, historical interactions, and the static environment. Through weighted fusion, it achieves comprehensive and multi-factor quantification of individual total risk, overcoming the limitations of traditional methods that are one-sided and static. This provides data reference for precise and personalized proactive safety warnings and interventions, thereby enhancing safety management capabilities in high-risk work environments. This invention achieves multi-dimensional risk assessment by dynamically combining the total risk value and its changing trend. Its advantages lie in not only classifying risks based on static thresholds but also introducing a rate-of-change parameter. This allows for early triggering of high-level warnings when the risk value is rapidly increasing even before reaching a high level, enhancing the predictive capability of the warnings. Simultaneously, the three-level warning system features differentiated audiovisual and tactile feedback, progressively strengthening from mild alerts to strong alarms, ensuring accurate information delivery and effective differentiation of urgency. In critical situations, an alarm containing location information is automatically sent to the backend, forming a closed-loop linkage from the field to monitoring, improving emergency response speed and collaborative handling efficiency. Attached Figure Description

[0016] The invention will now be further described with reference to the accompanying drawings.

[0017] Figure 1 This is a schematic diagram of the system described in this invention; Figure 2 This is a flowchart illustrating the method described in Embodiment 3 of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] like Figure 1 As shown, this application provides an intelligent management system for construction safety based on artificial intelligence; As an embodiment 1 of this application, it specifically includes: The multi-source spatiotemporal alignment acquisition module, based on the smart terminal worn by the operator, collects the operator's associated body motion data, spatial positioning data and environmental interaction data, and performs spatiotemporal alignment to construct a multi-source data set; The multi-scale feature fusion and extraction module, based on the constructed multi-source dataset, simultaneously performs micro-action feature extraction, macro-behavioral feature extraction, and environmental interaction feature extraction to generate a multi-scale feature set. The instantaneous risk component calculation module, based on a multi-scale feature set, calculates the current associated risk components of imbalance and fall, fall from height, and collision injury for workers, and comprehensively assesses the instantaneous comprehensive risk value of workers. The spatiotemporal interactive risk integration module constructs a risk attenuation function bound to the operator, assesses the interactive risk value of the operator in three-dimensional space, and calculates the total risk value of the operator by combining the static environmental risk value and the instantaneous comprehensive risk value. The tiered early warning push module executes tiered early warning operations based on the total risk value and its changing trend, and notifies the operators of the early warning notification through smart terminals. Example

[0020] An artificial intelligence-based intelligent management method for construction safety, comprising the following: This method: An intelligent management method for construction safety based on artificial intelligence, mainly includes 5 steps, which correspond to the 5 modules of the system described in this application, as follows: Step one: First, using the smart terminals worn by the workers, collect their associated body motion data, spatial positioning data, and environmental interaction data, and perform spatiotemporal alignment to construct multi-source data fusion. Specifically: The intelligent terminal is a wearable intelligent device pre-built by the operator. It is worn by the operator after arriving at the construction site and can perceive the operator's status and environment in real time and in all directions. In addition, it is equipped with sound and light alarm function and positioning device. It can locate the operator's current location in real time through the construction of a pre-built three-dimensional panoramic model of the construction site. The motion data of the body is collected based on the nine-axis inertial measurement unit integrated in the smart terminal, including three-axis acceleration ACC(t), three-axis angular velocity GYRO(t) and three-axis magnetic field strength MAG(t), where (t) indicates that it is related to time, and all subsequent data are related to time, rather than a single data point; For example, the triaxial acceleration ACC(t) can represent a data stream that includes the triaxial (x-axis, y-axis, and z-axis) accelerations at each time step.

[0021] Spatial positioning data is collected based on the UWB ultra-wideband positioning unit and inertial navigation module integrated into the smart terminal, including three-dimensional coordinates [X(t), Y(t), Z(t)]. The three-dimensional coordinates also interact with the three-dimensional panoramic model of the construction site in real time. For example, if the three-dimensional coordinates of the worker are (50, 30, 8.5), and the height displayed in the three-dimensional panoramic model of the construction site is 8.5 meters, then based on the three-dimensional panoramic model of the construction site, it can be known that the worker's current location is likely on the 3rd floor.

[0022] The environmental interaction data is collected based on the millimeter-wave radar integrated into the smart terminal, including the relative distance D_obj(t), relative velocity V_obj(t), and azimuth angle θ_obj(t) of obstacles within a 5-meter distance centered on the operator; It is important to note that the body motion data, spatial positioning data, and environmental interaction data are all collected based on a uniform sampling time interval preset by the operator. In other words, the body motion data, spatial positioning data, and environmental interaction data are synchronized in time.

[0023] The multi-source data set mainly includes body motion data, spatial positioning data, and environmental interaction data. The three types of data are time-aligned based on their respective bound time (i.e., acquisition time). The time-aligned body motion data, spatial positioning data, and environmental interaction data are converted into a pre-built 3D panoramic model of the construction site and mapped to the body coordinate system with the centroid of the worker as the origin. Extract the body motion data, spatial positioning data, and environmental interaction data from the body coordinate system, and combine them into a multi-source dataset. Step two: Based on the constructed multi-source data set, simultaneously perform micro-action feature extraction, macro-behavioral feature extraction, and environmental interaction feature extraction to generate a multi-scale feature set, that is, extract micro-action features, macro-behavioral features, and environmental interaction features from the multi-source data set.

[0024] Step 3: Based on the multi-scale feature set, calculate the current associated components of the worker's risk of fall due to imbalance, risk of fall from height, and risk of collision injury, and comprehensively assess the worker's instantaneous comprehensive risk value. Specifically: This step aims to transform the multi-scale feature set into specific and quantifiable risk values, mainly including the risk components of imbalance and fall, fall from height, and collision injury. The calculation principles for these three components are as follows: The imbalance fall risk component is mainly calculated from micro-movement characteristics. The higher the value of the imbalance fall risk component, the greater the likelihood of falling. The risk component of falling from height is determined by both macroscopic behavioral characteristics and spatial positioning (Z-coordinate height). For example, walking on the edge of a building increases the risk component of falling from height. The collision injury risk component is calculated based on environmental interaction characteristics. For example, if an object (such as a vehicle, or a worker autonomously approaching certain construction equipment) approaches the worker at high speed, the collision injury risk component increases.

[0025] Next, the three components are fused based on a predefined algorithm to obtain an instantaneous comprehensive risk value, which changes with time and with the current location of the operator.

[0026] Step 4: Construct a risk attenuation function that is tied to the workers, assess the interactive risk value of the workers in three-dimensional space, and calculate the total risk value of the workers by combining the static risk value of the environment and the instantaneous comprehensive risk value.

[0027] Step 5: Finally, based on the total risk value and its changing trend, a tiered early warning operation is executed, and the early warning notification is sent to the operators via smart terminals. Example

[0028] This embodiment, based on Embodiment 2, further discloses a method for generating a multi-scale feature set and calculating three components, thereby comprehensively evaluating the instantaneous integrated risk value. Figure 2 As shown, it specifically includes the following: First, as described in Example 2, based on the constructed multi-source dataset, micro-action feature extraction, macro-behavioral feature extraction, and environmental interaction feature extraction are performed simultaneously to generate a multi-scale feature set. The specific implementation method is as follows: First, there is the extraction of micro-motion features, which focuses on local and momentary instability of the body. The goal of this part is to quantify the subtle and rapid fluctuations in the body posture of workers, which are often precursors to slipping or falling. The acceleration vector and A_sum, i.e. the magnitude of the resultant acceleration, are calculated by using: A_sum=sqrt(ACC_x²+ACC_y²+ACC_z²), which represents the total intensity of the worker's body movement. Next, the standard deviation σ_A and kurtosis Kurt_A of the acceleration vector and A_sum within the operator's preset time window T1 are calculated. The larger the standard deviation σ_A, the more drastic the acceleration change and the more unstable the action of the operator during this period. The higher the kurtosis Kurt_A, the more extreme values ​​there are in the data. For example, an abnormally sharp acceleration peak indicates the moment when the foot suddenly slips.

[0029] Next, based on existing technology: the quaternion method (an efficient attitude calculation method without singularities) extracts the real-time pitch angle α_pitch(t) and roll angle α_roll(t) of the worker's torso relative to the ground from the triaxial acceleration ACC(t), triaxial angular velocity GYRO(t), and triaxial magnetic field strength MAG(t). Then, the attitude stability index S_posture is calculated using: S_posture=1 / (1+σ_α_pitch+σ_α_roll), where σ_α is the standard deviation of the angle within 1 second, σ_α_pitch represents the standard deviation of the real-time pitch angle within 1 second, and σ_α_roll represents the standard deviation of the roll angle within 1 second. The larger the value of (1+σ_α_pitch+σ_α_roll), the more stable the operator's body is, and vice versa.

[0030] For example, when a tired worker is carrying a heavy load, his upper body may sway from side to side, causing the standard deviation of α_roll, σ_α_roll, to remain high, thereby reducing S_posture and indicating that he is in an unstable state.

[0031] Next, we perform macro-level behavioral feature extraction, as follows: The three-dimensional coordinates [X(t), Y(t), Z(t)] of the workers are extracted in real time, and the movement trajectory of the workers within the time window T1 can be obtained by fitting the points on the three-dimensional panoramic model of the construction site. The curvature change rate of the movement trajectory C_curve(t)=|Δθ_heading / ΔS| is calculated, where Δθ_heading is the change in heading angle and ΔS is the movement arc length, both of which can be obtained from the movement trajectory.

[0032] Then, the motion entropy H_motion of the worker in three-dimensional space is calculated using H_motion=-Σ(p_i*log2(p_i)), where p_i is the probability of the velocity vector falling in each discrete direction interval on the unit sphere. That is, the direction of the worker's three-dimensional velocity vector is imagined as a point pointing on the sphere and divided into multiple regions. The probability p_i of the velocity direction falling in each region is statistically analyzed. Motion entropy H_motion is used to quantify the degree of disorder or unpredictability of the movement direction of workers. The higher the value of motion entropy H_motion, the more chaotic and irregular the movement direction of workers is. For example, workers walking normally on a flat ground have basically the same direction, and the entropy value is low. Workers who move around, frequently turn, and stop in a chaotic construction site have high entropy values. High entropy is itself a risk indicator because it means that their behavior is unpredictable and they are prone to unexpected interactions with the environment.

[0033] Finally, environmental interaction features are extracted, as follows: Only when the millimeter-wave radar detects that D_obj(t) < the preset safe distance D_s, it indicates that the trigger is successful, and the approach danger coefficient γ_approach=(D_s-D_obj(t)) / D_s+(V_obj(t) / V_s) is calculated; Where V_s is the preset reference speed, which can be set to the normal walking speed of the operator; (D_s-D_obj(t)) / D_s is the distance term, which is larger the closer to the obstacle (maximum 1), used to linearly penalize the reduction in distance; and (V_obj(t) / V_s) is the speed term, which is larger the relative speed of the obstacle (that is, the faster the operator approaches the obstacle).

[0034] Finally, micro-motion features are extracted, including acceleration vector and A_sum, standard deviation σ_A, kurtosis Kurt_A, real-time pitch angle α_pitch(t), roll angle α_roll(t), and attitude stability index S_posture; Extract macroscopic behavioral features, including the rate of change of curvature C_curve(t) and the motion entropy H_motion; Extract environmental interaction features, including the proximity hazard factor γ_approach; The micro-action features, macro-behavioral features, and environmental interaction features are combined to generate a multi-scale feature set F(t).

[0035] The posture stability index S_posture is extracted from the multi-scale feature set F(t), and the posture stability index S_posture(t) of the operator over time is determined. Then, the standard deviation σ_A and kurtosis Kurt_A are extracted, and the standard deviation σ_A(t) and Kurt_A(t) over time are determined. use: R_stumble(t) = w1*(1-S_posture(t)) + w2*(σ_A(t) / σ_A_max) + w3*(|Kurt_A(t)-3| / Kurt_norm) calculates the imbalance fall risk component R_stumble(t); Where w1, w2, and w3 are preset weighting coefficients, and w1 + w2 + w3 = 1, σ_A_max is a preset acceleration standard deviation threshold, and Kurt_norm is a kurtosis reference value. The 3 in (|Kurt_A(t)-3| / Kurt_norm) represents the normal value of 3 represented by the normal distribution, which is used to measure the degree of acceleration kurtosis deviating from the normal value of 3. The greater the deviation, the more abnormal shocks there are and the greater the risk contribution. Operators can also set this value according to the actual situation. (1-S_posture(t)) is used to invert the attitude stability index S_posture(t). The lower S_posture (the less stable), the higher 1-S_posture, and the greater the risk contribution. σ_A(t) / σ_A_max is used to standardize acceleration fluctuations. The greater the fluctuation, the closer it is to or exceeds the threshold σ_A_max, the greater the contribution of this item.

[0036] Next, the fall risk component R_fall(t) is calculated based on the physical reality of the fall risk. R_fall(t)=f_height(Z(t))*[λ1*g_dist(D_edge(t))+λ2*(H_motion(t) / H_max)+λ3*(C_curve(t) / C_max)]; f_height(Z(t)) is the height penalty function, f_height(Z(t))=1+log(1+Z(t) / Z_ref), its value increases as the height coordinate Z(t) of the worker increases, because height is the fundamental premise of the risk of falling from height, and Z_ref is the preset reference height; D_edge(t) is the projected horizontal distance from the operator to the nearest danger edge, calculated in real time based on the operator's three-dimensional coordinates [X(t),Y(t),Z(t)] and the pre-built three-dimensional panoramic model of the construction site. The danger edge is predefined by the operator in conjunction with the three-dimensional panoramic model of the construction site, such as the edge of the floor slab, the edge of the opening, etc. g_dist(D_edge(t)) is the edge distance risk function; g_dist(D_edge(t))=exp(-D_edge(t) / D_safe); The value of g_dist(D_edge(t)) increases as D_edge(t) decreases. D_safe is a preset safe distance threshold. When the distance from the edge D_edge(t) is very small, this value rises sharply and approaches 1, which strongly amplifies the risk. This is the most direct risk factor. (H_motion(t) / H_max) indicates that when a worker is near the edge of danger, the more chaotic and unpredictable their behavior (high entropy value), the greater the probability of falling. (C_curve(t) / C_max) indicates that when a worker is near the edge of danger, a sudden turn or hesitation (large change in curvature) poses a higher risk and a greater possibility of falling. λ1, λ2, λ3 are preset weighting coefficients, and λ1+λ2+λ3=1. H_max and C_max are preset motion entropy threshold and curvature change rate threshold, respectively. For example, if a worker walks steadily in the middle of a 10-meter-high floor slab, then f_height is medium, but D_edge is large, g_dist is close to 0, H_motion and C_curve are also small, so R_fall is low; When the workers walk to 0.5 meters from the edge of the floor slab, f_height remains unchanged, but g_dist increases sharply to close to 1, and R_fall will become extremely high in an instant.

[0037] Next, the collision damage risk component R_collision(t) is calculated; R_collision(t)=max(γ_approach(t))*[1+η*(||V_worker(t)|| / V_worker_max)]; max(γ_approach(t)) represents the maximum value of the approach hazard coefficients of all obstacles calculated from millimeter-wave radar data at any time t, with the most dangerous obstacle determining the current collision risk; ||V_worker(t)|| is the velocity factor, which is the speed of the worker at time t calculated based on spatial positioning data. If the worker is stationary, ||V_worker||=0, the factor is 1, and the collision risk is entirely determined by the obstacle. If the worker is also moving quickly, the factor is greater than 1, amplifying the total risk because the relative kinetic energy is greater and the consequences of the collision are more severe. η is the operator's preset adjustment coefficient, which is used to adjust the intensity of the influence of their own speed. Finally, the instantaneous comprehensive risk value R_instant(t) is calculated by combining the imbalance fall risk component R_stumble(t), the fall risk component R_fall(t), and the collision injury risk component R_collision(t) with the formula R_instant(t) = β1*R_stumble(t) + β2*R_fall(t) + β3*R_collision(t). Among them, β1, β2, and β3 are preset comprehensive weight coefficients, and β1 + β2 + β3 = 1. The values ​​of comprehensive weight coefficients β1, β2, and β3 are different in different scenarios. For example, in high-altitude work areas, β2 should be set very high to ensure that the system is extremely sensitive to the risk of falling from heights. The specific value is determined by the operator. Example

[0038] This embodiment, based on embodiment 3, further discloses a method for constructing a risk attenuation function bound to the operator, assessing the operator's interactive risk value in three-dimensional space, and calculating the operator's total risk value by combining the static environmental risk value and the instantaneous comprehensive risk value. Specifically, it includes the following: Based on the actual conditions of the construction site, it can be seen that the risk to workers depends not only on the current moment, but also on their recent history of activity in dangerous areas. For example, if a worker has been on the dangerous edge of a high-rise building multiple times due to their own habits or work needs, then a comprehensive assessment should be made in conjunction with historical risk-related data. First, create a personal risk footprint profile for each operator, namely a personal historical risk location record list L_risk, and store it persistently. Each record in the personal historical risk location record list L_risk records the three-dimensional coordinates [X_risk, Y_risk, Z_risk] of the risk location when the worker's current location is at risk (the judgment of risk will be given later); The three-dimensional coordinates [X_risk, Y_risk, Z_risk] of the risk location correspond to and are bound to the initial fall risk component R_fall_init of the operator at this risk location (that is, the fall risk component R_fall at the three-dimensional coordinates [X_risk, Y_risk, Z_risk] of the risk location, named and distinguished by the initial R_fall_init) and the generation timestamp T_risk of the record generated by the three-dimensional coordinates [X_risk, Y_risk, Z_risk] of the risk location and the initial fall risk component R_fall_init.

[0039] If the instantaneous fall risk component R_fall(t) exceeds the preset activation threshold R_fall_th, then the worker's current spatial positioning data [X(t),Y(t),Z(t)] will be used as the three-dimensional coordinates of the risk location, i.e., the three-dimensional coordinates of the risk location [X_risk,Y_risk,Z_risk]. The fall risk component R_fall(t) calculated at the spatial positioning data [X(t),Y(t),Z(t)] will be used as the initial fall risk component R_fall_init associated with the three-dimensional coordinates of the risk location [X_risk,Y_risk,Z_risk]. This will form a new record with the current timestamp and be added to L_risk.

[0040] Next, the risk decay function Φ(Δt,R_init) bound to the operator is constructed to calculate the risk value of any risk location after decay over time, as follows: R_decayed=Φ(Δt,R_init)=R_init*exp(-k*Δt); Among them, the risk decay function Φ(Δt,R_init) is essentially an exponential decay model, where Δt is the time difference between the current assessment time and the risk record timestamp T_risk, R_init is the initial high fall risk component R_fall_init, the initial risk R_init will decrease as the time difference Δt increases, k is the preset decay coefficient, and exp() is the exponential function. Next, obtain the spatial location data [X(t),Y(t),Z(t)] of the worker at the current moment, and calculate the distance d_i = sqrt((X(t)-X_risk)²+(Y(t)-Y_risk)²+(Z(t)-Z_risk)²) between the worker and all historical risk locations in the worker's personal historical risk location record list L_risk, where d_i represents the distance between any record in the personal historical risk location record list L_risk and the spatial location data [X(t),Y(t),Z(t)]. Next, the personal historical risk location record list L_risk is traversed to obtain all records in it where the distance d_i between the operator's current position [X(t),Y(t),Z(t)] and the operator's current position [X(t),Y(t),Z(t)] is less than the preset interaction radius R_int. The attenuated risk value R_decayed associated with each record is calculated by using the risk decay function Φ(Δt,R_init). Obtain the maximum value among all risk values ​​R_decayed and define it as the interaction risk value R_interaction(t) of the operator at the current moment: If, after calculation, there is no record that satisfies d_i < R_int, then set R_interaction(t) = 0.

[0041] Then, obtain the current three-dimensional coordinates [X(t),Y(t),Z(t)] of the operator, and determine the environmental static risk value R_environment(t) corresponding to the three-dimensional coordinates [X(t),Y(t),Z(t)] based on the pre-constructed environmental static risk map of the operator. The environmental static risk map is a pre-drawn hazard map that is linked in real time with the three-dimensional panoramic model of the construction site. Based on the three-dimensional panoramic model of the construction site, safety experts or through analysis of historical accident data, based on the static environmental hazard sources, assign risk level numerical labels to different areas. For example, R_environment is higher at the edge of deep foundation pits, below high-voltage lines, and within the turning radius of large equipment, possibly ranging from 90 to 100. In hardened and leveled centralized processing areas and living areas, the R_environment value may be lower, possibly between 5 and 15.

[0042] Combining the instantaneous comprehensive risk value R_instant(t), the interaction risk value R_interaction(t), and the static environmental risk value R_environment(t), the total risk value R_total(t) for the workers is calculated: R_total(t)=γ1*R_instant(t)+γ2*R_interaction(t)+γ3*R_environment(t) Among them, γ1, γ2, and γ3 are preset fusion weight coefficients, and γ1+γ2+γ3=1. Operators can dynamically adjust them according to the worker's job type, experience level, and the specific area where the worker is located. Example

[0043] This embodiment, based on embodiment 4, further discloses a method for performing tiered early warning operations based on the total risk value and its changing trend, specifically including the following: This embodiment, as the final step in the above embodiments, obtains the determined total risk value and, based on the total risk value and its changing trend (i.e., the total risk value corresponding to different times), performs a tiered early warning operation, as follows: First, obtain the risk warning thresholds preset by the operator, including the first threshold Th_low, the second threshold Th_medium and the third threshold Th_high, where 0 < Th_low < Th_medium < Th_high ≤ 1, and [0,1] represents the risk warning threshold range. The first threshold Th_low, the second threshold Th_medium and the third threshold Th_high all belong to this range. Next, the rate of change V_risk(t) of the total risk value R_total(t) within the preset time window T2 is calculated, where the duration of T2 is greater than that of T1, meaning that one time window T2 is one monitoring cycle, and this process is repeated cyclically. Then, based on the threshold range to which the total risk value R_total(t) belongs and the rate of change V_risk(t) of the total risk value R_total(t) within the preset time window T2, the warning level is determined, including attention level, warning level, and critical level, as follows: If: R_total(t)≥Th_high (the absolute value of risk has exceeded the highest threshold, and the strongest measures must be taken immediately regardless of the trend) or R_total(t)≥Th_medium and V_risk(t)>the operator's preset rise rate threshold V_up_th (it means that although the absolute value of risk has not reached the highest, it is already in the high-risk range and is rapidly deteriorating, so an early warning must be issued), it is judged as a critical level warning; If Th_medium≤R_total(t)<Th_high and V_risk(t)≤V_up_th (the risk is high, but the current trend is stable or slowly rising without a sharp deterioration, a clear warning is issued, but it is not necessary to trigger the highest alert), or Th_low≤R_total(t)<Th_medium and V_risk(t)>V_up_th (the risk is at a low to medium level, but is rising sharply, the warning should be upgraded in advance before the risk exceeds the median line, giving operators more reaction time); If Th_low≤R_total(t)<Th_medium and V_risk(t)≤V_up_th, it means that the risk is in the range that needs attention and there are no signs of rapid deterioration, and it is judged as a warning level. If R_total(t) < Th_low, it means the workers are safe, and the situation is determined to be without warning. When a warning level alert is issued, a low-frequency intermittent vibration and a yellow visual alert are emitted through the smart terminal to remind workers to pay attention or conduct self-checks and correct minor unsafe behaviors. When a warning level alert is issued, a continuous vibration, a red visual alert, and a level one buzzer sound will be emitted through the smart terminal, requiring the operator to immediately stop the current dangerous action and actively investigate the source of the hazard. When a critical level warning is issued, a high-intensity continuous vibration, a flashing red visual alert, and a level-two sharp buzzer sound are emitted via the smart terminal, ordering the workers to take the most urgent evacuation actions. Simultaneously, an alarm message containing the workers' three-dimensional coordinates and the warning level is sent to the operators, initiating external rescue coordination. Some data in the formulas described above have been numerically calculated with dimensions removed. Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0044] The above description is merely an example and illustration of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

[0045] It should be stated that all user data collected in this application was collected with the user's consent and authorization. Furthermore, the uses of user data are legal and compliant, and the use and processing of user data comply with the relevant laws, regulations, and standards of the relevant regions.

Claims

1. A construction safety intelligent management method based on artificial intelligence, characterized in that, The method includes: Step 1: Based on the smart terminal worn by the workers, collect the workers' body motion data, spatial positioning data and environmental interaction data, and perform spatiotemporal alignment to construct a multi-source data set; Step 2: Based on the constructed multi-source dataset, simultaneously perform micro-action feature extraction, macro-behavioral feature extraction, and environmental interaction feature extraction to generate a multi-scale feature set; Step 3: Based on the multi-scale feature set, calculate the current associated components of the worker's risk of imbalance and fall, risk of fall from height, and risk of collision injury, and comprehensively assess the worker's instantaneous comprehensive risk value. Step 4: Construct a risk attenuation function that is tied to the workers, assess the interactive risk value of the workers in three-dimensional space, and calculate the total risk value of the workers by combining the static environmental risk value and the instantaneous comprehensive risk value. Step 5: Based on the total risk value and its changing trend, execute the tiered early warning operation and notify the operators of the early warning through the smart terminal.

2. The method according to claim 1, characterized in that, In step one, the motion data of the body is collected based on the nine-axis inertial measurement unit integrated in the smart terminal, including three-axis acceleration ACC(t), three-axis angular velocity GYRO(t) and three-axis magnetic field strength MAG(t), where (t) represents the time relationship; Spatial positioning data is collected based on the UWB ultra-wideband positioning unit and inertial navigation module integrated in the smart terminal, including three-dimensional coordinates [X(t), Y(t), Z(t)]. The environmental interaction data is collected based on the millimeter-wave radar integrated into the smart terminal, including the relative distance D_obj(t), relative velocity V_obj(t), and azimuth angle θ_obj(t) of obstacles within a 5-meter distance centered on the operator; Body motion data, spatial positioning data, and environmental interaction data are collected based on a preset uniform sampling time interval.

3. The method according to claim 2, characterized in that, In step one, the specific method for constructing the multi-source data set is as follows: Time alignment is performed based on the time points bound to the body motion data, spatial positioning data, and environmental interaction data, respectively. The time-aligned body motion data, spatial positioning data, and environmental interaction data are converted into a pre-built 3D panoramic model of the construction site and mapped to the body coordinate system with the centroid of the worker as the origin. The body motion data, spatial positioning data, and environmental interaction data in the body coordinate system are extracted and combined into a multi-source data set.

4. The method according to claim 3, characterized in that, In step two, the specific method for generating the multi-scale feature set is as follows: S1, Micro-motion feature extraction: S11, calculate the standard deviation σ_A and kurtosis Kurt_A of the acceleration vector sum A_sum=sqrt(ACC_x²+ACC_y²+ACC_z²) within the time window T1, where sqrt() is the square root function, and the time window T1 is the time period preset by the operator; S12, calculate the real-time pitch angle α_pitch(t) and roll angle α_roll(t) of the worker's torso relative to the ground based on the quaternion method; S13, calculate the attitude stability index S_posture=1 / (1+σ_α_pitch+σ_α_roll), where σ_α is the standard deviation of the angle within 1 second, σ_α_pitch represents the standard deviation of the real-time pitch angle within 1 second, and σ_α_roll represents the standard deviation of the roll angle within 1 second; S2, Macro-behavioral feature extraction: S21. Based on the three-dimensional coordinates [X(t),Y(t),Z(t)] of the operator, determine the movement trajectory of the operator within the time window T1, and calculate the rate of change of curvature of the movement trajectory: C_curve(t)=|Δθ_heading / ΔS|, where Δθ_heading is the change in heading angle and ΔS is the movement arc length; S22, calculate the motion entropy H_motion of the worker in three-dimensional space = -Σ(p_i*log2(p_i)), where p_i is the probability of the velocity vector in each discrete direction interval on the unit sphere; S3, Environmental Interaction Feature Extraction: S31, when the millimeter-wave radar detects that D_obj(t) < the preset safe distance D_s, the following is adopted: γ_approach=(D_s-D_obj(t)) / D_s+(V_obj(t) / V_s); Calculate the approach hazard factor γ_approach, where V_s is the preset reference speed; Micro-motion features are extracted, including acceleration vector sum A_sum, standard deviation σ_A, kurtosis Kurt_A, real-time pitch angle α_pitch(t), roll angle α_roll(t), and attitude stability index S_posture; Extract macroscopic behavioral features, including the rate of change of curvature C_curve(t) and the motion entropy H_motion; Extract environmental interaction features, including the proximity hazard factor γ_approach; The micro-action features, macro-behavioral features, and environmental interaction features are combined to generate a multi-scale feature set F(t).

5. The method according to claim 4, characterized in that, In step three, the specific method for comprehensively assessing the instantaneous comprehensive risk value of the operators is as follows: S51, Calculate the imbalance fall risk component R_stumble(t): R_stumble(t)=w1*(1-S_posture(t))+w2*(σ_A(t) / σ_A_max)+w3*(|Kurt_A(t)-3| / Kurt_norm); Where w1, w2, w3 are preset weight coefficients, and w1+w2+w3=1, σ_A_max is a preset acceleration standard deviation threshold, and Kurt_norm is a kurtosis reference value; S52, Calculate the fall risk component R_fall(t): R_fall(t)=f_height(Z(t))*[λ1*g_dist(D_edge(t))+λ2*(H_motion(t) / H_max)+λ3*(C_curve(t) / C_max)]; Where f_height(Z(t)) is the height penalty function, f_height(Z(t))=1+log(1+Z(t) / Z_ref), and Z_ref is the preset reference height; D_edge(t) is the projected horizontal distance from the worker to the nearest danger edge, calculated in real time based on the worker's three-dimensional coordinates [X(t),Y(t),Z(t)] and a pre-built three-dimensional panoramic model of the construction site. g_dist(D_edge(t)) is the edge distance risk function, g_dist(D_edge(t))=exp(-D_edge(t) / D_safe), where D_safe is the preset safe distance threshold; λ1, λ2, λ3 are preset weighting coefficients, and λ1+λ2+λ3=1. H_max and C_max are preset motion entropy threshold and curvature change rate threshold, respectively. S53, Calculate the collision damage risk component R_collision(t): R_collision(t)=max(γ_approach(t))*[1+η*(||V_worker(t)|| / V_worker_max)]; Where max(γ_approach(t)) is the maximum value of the approach hazard coefficients of all obstacles calculated from millimeter-wave radar data at time t, ||V_worker(t)|| is the speed of the worker at time t calculated based on spatial positioning data, V_worker_max is the preset safe speed limit for the worker, and η is the preset adjustment coefficient. S54, based on the imbalance fall risk component R_stumble(t), the height fall risk component R_fall(t), and the collision injury risk component R_collision(t), calculate the instantaneous comprehensive risk value R_instant(t): R_instant(t)=β1*R_stumble(t)+β2*R_fall(t)+β3*R_collision(t); Wherein, β1, β2, and β3 are preset comprehensive weight coefficients, and β1 + β2 + β3 = 1.

6. The method according to claim 5, characterized in that, In step four, the risk attenuation function bound to the operator is constructed, and the specific method for assessing the interactive risk value of the operator in three-dimensional space is as follows: Create a personal historical risk location record list L_risk for each operator, where any record contains the three-dimensional coordinates of the risk location [X_risk, Y_risk, Z_risk]. The three-dimensional coordinates of the risk location correspond to the initial fall risk component R_fall_init and the record generation timestamp T_risk; When the instantaneous fall risk component R_fall(t) exceeds the preset activation threshold R_fall_th, the spatial positioning data [X(t),Y(t),Z(t)] of the worker at the current moment is used as the three-dimensional coordinates of the risk location, and R_fall(t) is used as the initial fall risk component R_fall_init. Together with the current timestamp, a new record is formed and added to L_risk. Construct the risk decay function Φ(Δt,R_init) that is bound to the operator, and calculate the risk value of any risk location after the three-dimensional coordinates decay over time: R_decayed=Φ(Δt,R_init)=R_init*exp(-k*Δt) Where Δt is the time difference between the current assessment time and the risk record timestamp T_risk, R_init is the initial high-altitude fall risk component R_fall_init, k is the preset attenuation coefficient, and exp() is the exponential function; Based on the spatial positioning data of the operator at the current moment [X(t),Y(t),Z(t)], calculate the distance d_i to all historical risk locations in L_risk = sqrt((X(t)-X_risk)²+(Y(t)-Y_risk)²+(Z(t)-Z_risk)²); Iterate through L_risk, and for each record whose distance from the current position d_i is less than the preset interaction radius R_int, calculate its decayed risk value R_decayed based on the risk decay function Φ(Δt,R_init); Define the maximum value among all risk values ​​R_decayed as the interaction risk value R_interaction(t) of the operator at the current moment: If there is no record that satisfies d_i < R_int, let R_interaction(t) = 0.

7. The method according to claim 6, characterized in that, In step four, the specific method for calculating the total risk value of the workers by combining the static environmental risk value and the instantaneous comprehensive risk value is as follows: S71, based on the operator's current three-dimensional coordinates [X(t),Y(t),Z(t)], query the pre-constructed environmental static risk map of the operator and obtain the environmental static risk value R_environment(t) corresponding to the three-dimensional coordinates; Among them, the environmental static risk map is constructed based on the three-dimensional panoramic model of the construction site, and predefines the risk level values ​​related to static environmental hazards for different spatial areas in the model. S72, combining the instantaneous comprehensive risk value R_instant(t), the interaction risk value R_interaction(t), and the static environmental risk value R_environment(t), calculate the total risk value R_total(t) for the workers: R_total(t)=γ1*R_instant(t)+γ2*R_interaction(t)+γ3*R_environment(t) Wherein, γ1, γ2, and γ3 are preset fusion weight coefficients, and γ1+γ2+γ3=1.

8. The method according to claim 7, characterized in that, In step five, the specific method for performing tiered early warning operations based on the total risk value and its changing trend is as follows: S81, extract the risk warning threshold, including the first threshold Th_low, the second threshold Th_medium and the third threshold Th_high, where 0 < Th_low < Th_medium < Th_high ≤ 1, and [0,1] represents the risk warning threshold range. S82, calculate the rate of change V_risk(t) of the total risk value R_total(t) within a preset time window T2, where the duration of T2 is greater than that of T1; S83, based on the threshold range to which the total risk value R_total(t) belongs and its rate of change V_risk(t), determine the warning level, including attention level, warning level, and critical level: If R_total(t)≥Th_high or R_total(t)≥Th_medium and V_risk(t)>preset rise rate threshold V_up_th, it is judged as a critical level warning; If Th_medium≤R_total(t)<Th_high and V_risk(t)≤V_up_th, or Th_low≤R_total(t)<Th_medium and V_risk(t)>V_up_th, it is determined to be a warning level alert; If Th_low≤R_total(t)<Th_medium and V_risk(t)≤V_up_th, it is determined to be a warning at the attention level; If R_total(t) < Th_low, then it is determined to be a state without warning; When S84 is determined to be a warning level, a low-frequency intermittent vibration and a yellow visual alert are emitted through the smart terminal. When the warning level is determined, a continuous vibration, a red visual alert, and a level one buzzer sound will be emitted through the smart terminal. When a critical level warning is issued, the system will send a high-intensity continuous vibration, a flashing red visual alert, and a level two sharp buzzer sound through the smart terminal, and simultaneously send alarm information including the three-dimensional coordinates of the operator and the warning level to the operator.

9. A construction safety intelligent management system based on artificial intelligence, characterized in that, The system includes: The multi-source spatiotemporal alignment acquisition module, based on the smart terminal worn by the operator, collects the operator's associated body motion data, spatial positioning data and environmental interaction data, and performs spatiotemporal alignment to construct a multi-source data set; The multi-scale feature fusion and extraction module, based on the constructed multi-source dataset, simultaneously performs micro-action feature extraction, macro-behavioral feature extraction, and environmental interaction feature extraction to generate a multi-scale feature set. The instantaneous risk component calculation module, based on a multi-scale feature set, calculates the current associated risk components of imbalance and fall, fall from height, and collision injury for workers, and comprehensively assesses the instantaneous comprehensive risk value of workers. The spatiotemporal interactive risk integration module constructs a risk attenuation function bound to the operator, assesses the interactive risk value of the operator in three-dimensional space, and calculates the total risk value of the operator by combining the static environmental risk value and the instantaneous comprehensive risk value. The tiered early warning push module executes tiered early warning operations based on the total risk value and its changing trend, and notifies the operators of the early warning notifications via smart terminals.

Images