A construction emergency early warning method and system

Abstract

This invention discloses a construction emergency early warning method and system. The method includes the following steps: collecting the three-dimensional coordinates of construction workers, using a sliding time window combined with indicators such as wavelet packet energy entropy, and generating a multi-scale motion disorder index through principal component analysis; identifying continuously unstable personnel based on the disorder index, segmenting their trajectories, and constructing a weighted graph by combining Hidden Markov Model and Building Information Model environmental parameters to calculate cognitive mismatch; mapping the cognitive mismatch to a spatial grid, calculating the local Moran index, and fusing density gradient and mechanical operation sequence resonance results to construct a propagation weight matrix; guiding the construction of a heterogeneous graph based on the weight matrix, calculating the risk impact propagation potential energy, and coupling historical disorder sequence analysis and relative entropy to obtain a group-level instability prediction value; for individuals whose prediction values ​​exceed the limit, outputting a comprehensive crisis level through motion trend prediction, spatial intersection, and shortest path algorithms. This invention improves the safety of construction operations.

Description

Technical Field

[0001] This invention relates to the field of construction, and more particularly to a method and system for early warning of construction emergencies. Background Technology

[0002] With the continuous advancement of urbanization and rapid development of infrastructure construction in my country, complex projects such as high-rise buildings, underground spaces, and large-scale complexes are increasingly common. Construction sites are characterized by complex spatial structures, densely overlapping work surfaces, multi-trade collaboration, and high-frequency operation of machinery and equipment, leading to highly dynamic and concealed safety risks. Traditional construction safety management mainly relies on passive, experience-driven methods such as manual inspections, safety training, warning signs, and video surveillance, which have many insurmountable limitations. Meanwhile, construction sites, as typical unstructured indoor environments, suffer from severe GPS signal attenuation or even complete failure, making it impossible to directly apply mature outdoor positioning technologies. In recent years, UWB indoor positioning technology has been gradually introduced to construction sites, but most systems remain at the level of location visualization, failing to transform location data into safety early warning knowledge. Based on this, this invention proposes a construction emergency early warning method and system. Summary of the Invention

[0003] This invention provides a construction emergency early warning method, characterized in that it includes:

[0004] S10. Based on high-frequency three-dimensional coordinate data collected by UWB base stations, a multi-scale motion disorder index is generated by combining wavelet packet energy entropy, directional autocorrelation attenuation ratio and velocity change rate quantile through a sliding time window and using principal component analysis.

[0005] S20. Identify individuals with persistent instability based on the multi-scale motion disorder index, perform Bayesian variable point segmentation on their trajectories, construct a weighted graph by combining the state transition spectrum radius of the Hidden Markov Model with the environmental parameters extracted from the Building Information Model, calculate the perceptual load integral, and generate a cognitive mismatch index.

[0006] S30. Map cognitive mismatch to a spatial grid, calculate local Moran index to identify risk clusters, fuse density gradient and mechanical operation sequence resonance results, and construct a dynamic disturbance propagation weight matrix to characterize the spatial impact of external disturbances on risk diffusion.

[0007] S40. Based on the dynamic interference propagation weight matrix, construct a heterogeneous graph of personnel, use improved webpage ranking to calculate the propagation potential of risk influence, and perform symbolic time series analysis and Markov analysis on its historical disorder sequence. Use relative entropy to measure the intensity of behavioral inertia locking, and couple the two to obtain the predicted value of group-level instability.

[0008] S50. For individuals whose predicted values ​​exceed the limits, the comprehensive urgency level is output as the basis for graded early warning through motion trend prediction algorithm, spatial intersection detection algorithm and shortest path algorithm.

[0009] The construction hazard early warning method described above involves using high-frequency three-dimensional coordinate data collected by UWB base stations. This data is then combined with wavelet packet energy entropy, directional autocorrelation attenuation ratio, and velocity change rate quantiles via a sliding time window, and principal component analysis is used to generate a multi-scale motion turbulence index. Specifically, this method comprises the following sub-steps:

[0010] Trajectory data is formed by collecting three-dimensional coordinates based on UWB positioning tags. Based on the trajectory data, wavelet packet decomposition of vertical acceleration is performed through a sliding window, and energy distribution entropy is calculated to quantify the degree of vertical motion disorder.

[0011] Analyze the horizontal change characteristics, standardize them by combining the vertical fluctuation entropy, and extract the instantaneous disorder component by fusing multi-dimensional motion characteristics through principal component analysis.

[0012] The Hearst exponent is calculated by sliding the transient disorder component sequence. If it remains above the threshold, it is determined to be a persistent instability. The multi-scale motion disorder index is output and a deep assessment is triggered.

[0013] The construction emergency early warning method described above involves identifying continuously unstable personnel based on a multi-scale motion disorder index, performing Bayesian variable point segmentation on their trajectories, constructing a weighted graph by combining the state transition spectrum radius of a Hidden Markov Model with environmental parameters extracted from a Building Information Model, calculating the perceived load integral, and generating a cognitive mismatch index. Specifically, it consists of the following sub-steps:

[0014] For individuals marked as persistently unstable, their trajectory data is retrieved and a Bayesian change point detection algorithm is applied to identify abrupt changes in motion patterns, dividing the trajectory into multiple behavioral segments;

[0015] Each behavioral segment is discretized into spatial nodes to construct a path weighted graph, and a hidden Markov model is fitted to extract behavioral predictability. Combined with environmental parameters analyzed by building information model, a comprehensive environmental pressure index is constructed.

[0016] The total perceived load is calculated using the trajectory-environment coupling curvature integral function, and the cognitive mismatch is calculated based on the total perceived load.

[0017] The construction emergency early warning method described above involves calculating the total perceived load using a trajectory-environment coupled curvature integral function, and then calculating the cognitive mismatch degree based on the total perceived load. Specifically, it comprises the following sub-steps:

[0018] Parameters are initialized based on the actual trajectory from the start to the end of the behavior segment, and weight values ​​are dynamically adjusted based on the semantic tags of the building information model.

[0019] The total sensed load on the path segment is obtained by numerically integrating the trajectory-environment coupling curvature integral function along the trajectory.

[0020] Select the segment with the lowest predictability among all behavioral segments, and generate the cognitive mismatch degree based on its highest total perceptual load value according to the cognitive generation formula.

[0021] The construction hazard early warning method described above maps cognitive mismatch to a spatial grid, calculates local Moran's index to identify risk clusters, integrates density gradient and mechanical operation sequence resonance results, and constructs a dynamic disturbance propagation weight matrix to characterize the spatial impact of external disturbances on risk diffusion. Specifically, it consists of the following sub-steps:

[0022] The cognitive mismatch of construction workers is mapped to a spatial grid, and the local Moran index is calculated to identify high-risk spatial clustering areas;

[0023] Based on the cosine of the angle between the population density gradient and the carbon monoxide concentration gradient, a population guidance distortion coefficient is constructed to quantify the guidance deviation between environmental hazards and population movement trends.

[0024] The risk resonance intensity function is used to reveal the deep coupling mechanism between external mechanical disturbances and internal group disorder. The presence of resonance activates the dynamic disturbance propagation weight matrix.

[0025] The construction hazard early warning method described above involves constructing a heterogeneous personnel graph based on a dynamic interference propagation weight matrix, calculating the risk impact propagation potential energy using improved webpage ranking, and performing symbolic time series analysis and Markov analysis on its historical disorder sequences. Relative entropy is used to measure the behavioral inertia locking strength, and the two are coupled to obtain a group-level instability prediction value. Specifically, it consists of the following sub-steps:

[0026] Construct a dynamic heterogeneous graph to simulate and calculate the potential energy of risk impact propagation;

[0027] To quantify whether individual behavior is trapped in an inertial lock-in state, symbolic time series analysis is performed on the historical multiscale motion disorder index time series of individuals in the neighborhood of each node;

[0028] For each node marked as high-risk, its risk impact propagation potential and behavioral inertia locking strength are combined to generate a group-level instability prediction value.

[0029] The construction hazard early warning method described above, in which individuals whose predicted values ​​exceed limits are assessed using a motion trend prediction algorithm, a spatial intersection detection algorithm, and a shortest path algorithm to output a comprehensive hazard level as the basis for graded early warning, is specifically divided into the following sub-steps:

[0030] Based on the current motion state of high-risk individuals, a fan-shaped spatial envelope region to be occupied in the future is generated by introducing short-term trajectory extrapolation with directional deviation.

[0031] Calculate the proportion of the spatial intersection volume between the trajectory envelope and the hazard source buffer zone in the building information model to quantify its probability of intruding into the high-risk area;

[0032] By combining path complexity, risk propagation potential, and intersection, a weighted formula is used to generate a graded comprehensive hazard level to drive graded alarms.

[0033] This invention also provides a construction emergency early warning system, comprising:

[0034] Acquisition and generation module: Based on high-frequency three-dimensional coordinate data acquired by UWB base stations, a multi-scale motion disorder index is generated by combining wavelet packet energy entropy, directional autocorrelation attenuation ratio and velocity change rate quantile through a sliding time window and using principal component analysis.

[0035] Identification and Calculation Module: Identifies individuals with persistent instability based on the multi-scale motion disorder index, performs Bayesian variable point segmentation on their trajectories, constructs a weighted graph by combining the state transition spectrum radius of the Hidden Markov Model with environmental parameters extracted from the Building Information Model, calculates the perceptual load integral, and generates a cognitive mismatch index.

[0036] Mapping and Construction Module: Maps cognitive mismatch to a spatial grid, calculates local Moran index to identify risk clusters, integrates density gradient and mechanical operation sequence resonance results, and constructs a dynamic disturbance propagation weight matrix to characterize the spatial impact of external disturbances on risk diffusion;

[0037] Risk propagation and instability prediction module: Based on the dynamic interference propagation weight matrix, a heterogeneous graph of personnel is constructed. An improved webpage ranking is used to calculate the propagation potential of risk influence. Symbolic time series analysis and Markov analysis are performed on its historical disorder sequence. The relative entropy is used to measure the intensity of behavioral inertia locking. The two are coupled to obtain the group-level instability prediction value.

[0038] Criticality Level Module: For individuals whose predicted values ​​exceed the limits, a comprehensive criticality level is output as the basis for graded early warning through motion trend prediction algorithm, spatial intersection detection algorithm, and shortest path algorithm.

[0039] The beneficial effects achieved by this invention are as follows: By combining high-precision positioning technologies such as UWB and Building Information Modeling, and utilizing trajectory prediction and spatial analysis algorithms, this invention enables precise monitoring of personnel activities at construction sites and timely alarms for potential risks, thereby improving the safety of construction workers during operations. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0041] Figure 1 This is a flowchart of a construction emergency early warning method provided in Embodiment 1 of this application.

[0042] Figure 2 This is a schematic diagram of a construction emergency early warning system provided in Embodiment 2 of this application. Detailed Implementation

[0043] 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, not all, of the embodiments of the present invention. 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.

[0044] Example 1

[0045] like Figure 1 As shown, Embodiment 1 of this application provides a construction emergency early warning method, including:

[0046] S10. Based on high-frequency three-dimensional coordinate data collected by UWB base stations, a multi-scale motion disorder index is generated by combining wavelet packet energy entropy, directional autocorrelation attenuation ratio, and velocity change rate quantile through a sliding time window and principal component analysis.

[0047] S11. Collect three-dimensional coordinates to form trajectory data based on UWB positioning tags. Based on the trajectory data, perform wavelet packet decomposition on the vertical acceleration through a sliding window and calculate the energy distribution entropy to quantify the degree of vertical motion disorder.

[0048] The UWB base station array continuously acquires ultra-wideband pulse signals transmitted by UWB positioning tags worn by construction workers at a high frequency of 50Hz. Combined with positioning algorithms, the three-dimensional coordinate data of the tags is calculated, forming a high-resolution spatiotemporal trajectory stream. A 75% overlapping sliding time window is used to segment the original trajectory, ensuring continuous capture of dynamic behavioral characteristics and preservation of local details. For each segmented trajectory, the vertical displacement sequence along the Z-axis is extracted and differentially analyzed to obtain the acceleration sequence. Subsequently, a 4-level wavelet packet decomposition using the db4 wavelet basis is performed, dividing the signal energy into 16 frequency bands. The energy proportion of each frequency band is calculated, and a vertical fluctuation entropy index is constructed based on information entropy theory to quantify the dispersion of human body vibration energy distribution in the vertical direction. High-frequency micro-vibrations and low-frequency large-amplitude fluctuations both lead to energy dispersion and increased entropy, thus sensitively reflecting abnormal posture control.

[0049] S12. Analyze the horizontal change characteristics, standardize them by combining them with the vertical fluctuation entropy, and extract the instantaneous disorder components by fusing multi-dimensional motion characteristics through principal component analysis.

[0050] Within the horizontal plane, the velocity direction angle sequence for each trajectory segment is calculated, i.e., the angle between the motion direction and true north. Autocorrelation analysis is then performed on this angle sequence to obtain the autocorrelation function value after 5 frames of lag. The ratio of this value to the zero-lag autocorrelation value is then calculated to obtain the direction autocorrelation decay ratio. This ratio reflects the short-term predictability of the motion direction; a ratio close to 1 indicates directional stability, while rapid decay indicates frequent turning or swaying. Simultaneously, the 95th quantile of the velocity direction angle change rate sequence is calculated as the drasticness of directional change to capture extreme turning behavior. The vertical directional fluctuation entropy, direction autocorrelation decay ratio, and drasticness of directional change are Z-score standardized to eliminate dimensional differences before being input into principal component analysis. Utilizing its dimensionality reduction and weight adaptive characteristics, the score of the first principal component with the highest variance contribution is extracted as the instantaneous turbulence component for that time period, effectively fusing multidimensional motion anomaly features.

[0051] S13. Calculate the Hearst exponent by sliding the instantaneous disorder component sequence. If it continues to be higher than the threshold, it is determined to be a persistent instability. Output the multi-scale motion disorder index and trigger a deep assessment.

[0052] For the time series of transient disorder components, the Hearst exponent is calculated segment by segment using the sliding window method. An exponent value greater than 0.5 indicates positive persistence, suggesting the current disorder may continue; an exponent value greater than 0.8 indicates strong long-range correlation, indicating a continuous evolutionary trend in the disorder state. A criterion is set: if the Hearst exponent is greater than 0.8 for three consecutive non-overlapping windows, the individual is determined to be in a state of persistent instability, excluding misjudgments caused by transient disturbances such as turning or bending over. The final output is a multi-scale motion disorder index containing time series data and trend status, serving as a trigger condition for entering in-depth risk assessment, ensuring that subsequent environmental coupling analysis is performed only on truly high-risk individuals.

[0053] S20. Identify individuals with persistent instability based on the multi-scale motion disorder index, perform Bayesian variable point segmentation on their trajectories, construct a weighted graph by combining the state transition spectrum radius of the Hidden Markov Model with environmental parameters extracted from the Building Information Model, calculate the perceptual load integral, and generate a cognitive mismatch index.

[0054] S21. For individuals marked as persistently unstable, retrieve their trajectory data and apply the Bayesian change point detection algorithm to identify abrupt changes in motion patterns, dividing the trajectory into multiple behavior segments.

[0055] For construction workers identified as having persistent instability, the complete trajectory point set within the most recent time window is retrieved, and a Bayesian change point detection algorithm is used to identify the moment of sudden change in the motion pattern. This method constructs a multivariate observation model based on the velocity, acceleration, and orientation angle sequences of the trajectory, sets the number of potential change points with Poisson distribution as a prior assumption, and maximizes the confidence of trajectory structure changes segment by segment through posterior probability, accurately locating behavioral turning points such as sudden stops, sharp turns, or accelerated escape, thereby dividing the entire trajectory into several behavioral segments with stable motion patterns.

[0056] S22. Discretize each behavior segment into spatial nodes to construct a path weighted graph, and fit a hidden Markov model to extract behavior predictability. Combine the environmental parameters analyzed by the building information model to construct a comprehensive environmental pressure index.

[0057] For each behavior segment, its trajectory is discretized into a series of spatial nodes, and a path-weighted graph structure is constructed using these points as graph nodes. A three-state Hidden Markov Model is then built based on the behavior segment, with states corresponding to uniform movement, stationary observation, and anomalous swaying, respectively. The Baum-Welch algorithm is used to learn the transition probability matrix and emission probability matrix from the trajectory observation sequence. The feature distribution of each state is extracted from the emission probability matrix, and its spectral radius is calculated from the transition probability matrix as a predictability index of the behavior pattern of that segment.

[0058] Meanwhile, spatial environmental parameters of the path traversed by the behavior segment are analyzed from the building information model, including net height gradient, light intensity variation coefficient, ground friction coefficient distribution entropy, and total emergency exit sight obstruction angle. These parameters are then combined according to empirical weights into a comprehensive environmental pressure index to quantify the cognitive load of the local work environment.

[0059] S23. Calculate the total perceived load using the trajectory-environment coupling curvature integral function, and calculate the cognitive mismatch based on the total perceived load.

[0060] The specific formula for the trajectory-environment coupled curvature integral function is as follows: ,in, For personnel trajectory parameterization curves, The trajectory curvature reflects abrupt changes in direction. It is the attenuation constant, used to control the sensitivity to changes in net height. This represents the change in net height along the path. Let be the illumination variation coefficient at that point. The gradient of the ground friction coefficient, These are the weighting coefficients.

[0061] The process of calculating the total perceived load is as follows: initializing parameters based on the actual trajectory from the start to the end of the behavior segment; and dynamically adjusting parameters based on the semantic tags of the building information model. Value; the total sensed load on the path segment is obtained by numerically integrating the trajectory-environment coupling curvature integral function along the trajectory.

[0062] Select the segment with the lowest predictability among all behavior segments, and use its highest total perceived load value. Based on the formula Generate cognitive mismatch, Let be the spectral radius of the state transition matrix for this behavior segment.

[0063] S30. Map cognitive mismatch to a spatial grid, calculate local Moran index to identify risk clusters, fuse density gradient and mechanical operation sequence resonance results, and construct a dynamic disturbance propagation weight matrix to characterize the spatial impact of external disturbances on risk diffusion.

[0064] S31. Map the cognitive mismatch of construction workers to a spatial grid and calculate the local Moran index to identify high-risk spatial clustering areas.

[0065] The cognitive mismatch degrees calculated in real time by all construction workers are used as individual attribute values. These values ​​are then mapped to a unified 0.5m × 0.5m resolution grid system on the construction site using spatial interpolation methods, forming a spatial risk field with grid cells as the basic unit. For each grid i, its local Moran index is calculated to identify high-risk spatial clustering patterns. ,in For the cognitive mismatch of grid i, The global mean. This represents the spatial connection strength between the i-th grid and the j-th grid. Let N be the cognitive mismatch degree of grid j, and N be the total number of grid cells. A spatial weight matrix is ​​constructed using a vehicle-type adjacency rule. The hypothesis test probability value p is calculated using the Monte Carlo permutation test. If a grid satisfies the condition that the local Moran index is greater than a set threshold and the hypothesis test probability value is less than a set threshold, it is determined to be a high-high type clustering core. This grid and its neighborhood are all high-risk and are marked as potential hotspots for cluster risk outbreaks for subsequent density and disturbance analysis.

[0066] S32. Based on the cosine of the angle between the population density gradient and the carbon monoxide concentration gradient, a population guidance distortion coefficient is constructed to quantify the guidance deviation between environmental hazards and population movement trends.

[0067] Spatially smoothing is applied to the current locations of all individuals using Gaussian kernel density estimation, with the kernel function bandwidth set to an empirically optimal value, generating a continuous crowd density field that reflects the degree of crowding in different areas. The gradient vector field of this density field is further calculated to represent the direction and intensity of crowd flow trends. Simultaneously, carbon monoxide concentration distribution is obtained from environmental monitoring nodes, and a carbon monoxide concentration field is constructed and its gradient is calculated. The cosine of the angle between the two gradient directions is then determined. Define the group guidance distortion coefficient as This indicator reflects the degree of deviation between population movement trends and the direction of harmful environmental spread. This represents the gradient vector field of population density.

[0068] S33. The risk resonance intensity function is used to reveal the deep coupling mechanism between external mechanical disturbances and internal group disorder. The existence of resonance activates the dynamic disturbance propagation weight matrix.

[0069] The specific formula is as follows: Where T is the sliding time window. Let be the crowd density field, representing the spatial location at time t. The number of people per unit area in the vicinity The Laplace expression for population density shows that negative values ​​correspond to density peaks or aggregation vortices, which are structural markers of population instability. The time derivative of the start-stop signals of heavy machinery such as tower cranes and pile drivers extracted from the device IoT logs is used to generate pulse signals at the moment of start-stop, which characterizes the instantaneous impact of external disturbances. The spectral coherence distance between the mechanical working cycle and the dominant frequency of crowd movement. This is an empirical scale parameter. If... If the value is greater than a set threshold, a significant resonance is determined to exist, and the dynamic interference propagation weight matrix is ​​activated. Its edge weights are designed as follows: ,in, Let i be the propagation weight from node i to node j. Let i be the Euclidean distance between grid cells i and j. For distance attenuation scale parameters, As a local sensitivity enhancement term, it transforms local guidance failure rate into propagation gain, making the risk more likely to spread to areas of cognitive mismatch and guidance failure. This is the distortion coefficient amplification factor. Let be the group-guided distortion coefficient of node j.

[0070] S40. Based on the dynamic interference propagation weight matrix, a heterogeneous graph of personnel is constructed. An improved webpage ranking is used to calculate the propagation potential of risk influence. Symbolic time series analysis and Markov analysis are performed on its historical disorder sequence. Relative entropy is used to measure the intensity of behavioral inertia locking. The two are coupled to obtain the predicted value of group-level instability.

[0071] S41. Construct a dynamic heterogeneous graph to simulate and calculate the potential energy of risk impact propagation.

[0072] Using each construction worker as a graph node, two types of edges are established: physical edges (undirected edges representing potential behavioral impacts, including following, avoidance, and collisions, when the spatial distance between two workers is less than 3 meters), and logical edges (directed edges representing information transmission paths within the organizational structure, based on the collaborative relationships in the task management system).

[0073] A dynamic heterogeneous graph is formed, and the dynamic interference propagation weight matrix is ​​used as a global propagation situation guidance signal.

[0074] The dynamic evolution equation of risk propagation potential energy in heterogeneous graphs is used to accurately simulate the nonlinear propagation process of risk on dynamic heterogeneous graphs. The specific formula is as follows: ,in, This represents the potential energy of risk propagation at time step k+1 for node i. This represents the potential energy of risk propagation at time step k for node j. Let i be the damping factor, using the formula The value of the damping factor is adaptively adjusted according to the intensity of the disturbance resonance in the individual's environment, where 'a' is a fixed constant. The normalization coefficient is... For the intensity of risk resonance, Let be the edge weight from neighbor j to i in the dynamic interference propagation weight matrix. M is the total number of nodes. Let i be the set of all neighboring nodes of node i. It is the edge weight from node j to node i at time step k. It is the edge weight from node j to node m at time step k. For the propagation efficiency function, It is the information propagation delay from node j to node i at time step k. It is a time interval.

[0075] S42. To quantify whether an individual's behavior is trapped in an inertial lock-in state, a symbolic time series analysis is performed on the historical multi-scale motion disorder index time series of individuals in the neighborhood of each node.

[0076] The original time series is approximated and reduced in dimensionality through segmented aggregation and mapped to six discrete symbols, realizing the abstract transformation from continuous dynamics to a sequence of behavioral states. Subsequently, based on the symbol sequences of all neighborhood individuals, the state transition frequency is statistically analyzed to construct the Markov transition matrix M, and the steady-state distribution of the Markov chain is further solved. That is, satisfying and The long-term behavioral distribution. Based on this, relative entropy is used as the core metric to calculate... , A 6-dimensional uniform distribution represents an ideal benchmark of completely random and irregular behavior; this relative entropy quantifies the actual behavior distribution. Information deviation from the ideal random distribution: if behavior is diverse and disordered, ,but If behavior is highly concentrated in a few states, then This indicates that the behavior has been locked in by inertia.

[0077] S43. For each node marked as high-risk, integrate its risk influence propagation potential and behavioral inertia locking strength to generate a group-level instability prediction value.

[0078] The specific formula for generating the population-level instability prediction value is as follows: ,in, The Sigmoid function is used to propagate the potential energy of risk impact. Normalizing the propagation potential energy into the activation probability of risk amplification capability This characterizes the rigidity of neighborhood behavioral patterns; the product of these two factors reflects the synergistic risk effect of highly transmissible individuals in a low-elasticity group environment. (Group-level instability prediction value) The larger the value, the more likely the node is to become a trigger point for group-level instability, thereby triggering tiered early warning or proactive intervention measures.

[0079] S50. For individuals whose predicted values ​​exceed the limits, the comprehensive urgency level is output as the basis for graded early warning through motion trend prediction algorithm, spatial intersection detection algorithm and shortest path algorithm.

[0080] S51. Based on the current motion state of high-risk individuals, the future sector-shaped spatial envelope region is generated by introducing short-term trajectory extrapolation with directional deviation.

[0081] For individuals deemed high-risk due to a group-level instability prediction value exceeding a set threshold, an individual-level crisis assessment process is initiated. Based on their current real-time location, velocity vector, and direction of movement, a motion trend prediction algorithm is used to extrapolate their possible paths over the next 3-5 seconds. A ±15° directional deviation angle is introduced to account for uncertainties in personnel turning, forming a trajectory envelope region that unfolds in a fan shape with the current direction of movement as its central axis. This region represents the set of spaces the individual may occupy in the near future, possesses dynamic updating characteristics, and serves as a crucial bridge connecting behavioral risk and physical environmental hazards.

[0082] S52. Calculate the proportion of the spatial intersection volume between the trajectory envelope and the hazard source buffer zone in the building information model, and quantify its possibility of intruding into the high-risk area.

[0083] A spatial intersection detection algorithm is used to perform a 3D spatial overlay analysis between the trajectory envelope region and predefined static hazard source buffers in the Building Information Model (BIM). Hazard sources include the operating radius of tower cranes, the edge of deep foundation pits, the activity range of pile drivers, and temporary power facilities, with buffers set according to safety regulations. The intersection degree index is obtained by calculating the proportion of the spatial intersection volume between the trajectory envelope and these buffers to the total volume of the trajectory envelope: 0 if there is no overlap, and close to 1 if the individual completely enters the hazard zone. This value quantifies the probability and extent to which an individual intrudes into a high-risk physical space in the short term.

[0084] S53. Combining path complexity, risk propagation potential, and intersection, a weighted formula is used to generate a graded comprehensive hazard level to drive graded alarms.

[0085] To further assess an individual's evacuation capability in a complex construction site environment, a path complexity assessment algorithm is employed. Based on a passable grid map constructed using Building Information Modeling (BIM), the A* algorithm is used to calculate the optimal reachable path length from the individual's current location to the nearest safe exit, and this length is compared with the ideal straight-line distance to quantify path complexity. Finally, based on the intersection index, risk propagation potential, and path complexity, a weighted hazard level formula is constructed, outputting a comprehensive hazard level within the range [0,10]. A higher value indicates that the individual is closer to the hazard source, has a stronger risk propagation capability, and has a more complex escape route. This triggers a tiered alarm, achieving a precise and interpretable real-time safety response.

[0086] Example 2

[0087] like Figure 2 As shown, Embodiment 2 of this application provides a construction emergency early warning system, including:

[0088] The acquisition and generation module uses high-frequency three-dimensional coordinate data acquired by UWB base stations. It combines wavelet packet energy entropy, directional autocorrelation attenuation ratio, and velocity change rate quantiles through a sliding time window, and then uses principal component analysis to generate a multi-scale motion turbulence index. Specifically, it consists of the following sub-steps:

[0089] The acquisition and distribution entropy submodule: It acquires three-dimensional coordinates to form trajectory data based on UWB positioning tags, performs wavelet packet decomposition on vertical acceleration through a sliding window based on the trajectory data, and calculates energy distribution entropy to quantify the degree of vertical motion disorder.

[0090] Extracting the disorder component submodule: Analyzing the horizontal change characteristics, standardizing the process by combining it with the vertical fluctuation entropy, and extracting the instantaneous disorder component by fusing multi-dimensional motion characteristics through principal component analysis.

[0091] Motion Disorder Index Submodule: The Hurst exponent is calculated by sliding the instantaneous disorder component sequence. If it is consistently higher than the threshold, it is determined to be a persistent instability. The multi-scale motion disorder index is output and a deep assessment is triggered.

[0092] The identification and calculation module identifies individuals with persistent instability based on a multi-scale motion disorder index, performs Bayesian variable point segmentation on their trajectories, constructs a weighted graph by combining the state transition spectrum radius of a Hidden Markov Model with environmental parameters extracted from a Building Information Model, calculates the perceived load integral, and generates a cognitive mismatch index. Specifically, it consists of the following sub-steps:

[0093] The mutation detection and segmentation submodule: For personnel marked as persistently unstable, their trajectory data is retrieved and a Bayesian change point detection algorithm is applied to identify abrupt changes in motion patterns, dividing the trajectory into multiple behavior segments.

[0094] Environmental stress submodule: Discretize each behavior segment into spatial nodes to construct a path weighted graph, and fit a hidden Markov model to extract behavior predictability. Combine the environmental parameters analyzed by building information model to construct a comprehensive environmental stress index.

[0095] Load and cognitive mismatch submodule: The total perceived load is calculated using the trajectory-environment coupling curvature integral function, and the cognitive mismatch degree is calculated based on the total perceived load.

[0096] The mapping and construction module maps cognitive mismatch to a spatial grid, calculates local Moran's indices to identify risk clusters, fuses density gradients and mechanical operation sequence resonance results, and constructs a dynamic disturbance propagation weight matrix to characterize the spatial impact of external disturbances on risk diffusion. Specifically, it consists of the following sub-steps:

[0097] High-risk identification submodule: Map the cognitive mismatch of construction workers to a spatial grid and calculate the local Moran index to identify high-risk spatial clustering areas.

[0098] The distortion coefficient submodule constructs a population guidance distortion coefficient based on the cosine of the angle between the population density gradient and the carbon monoxide concentration gradient, quantifying the guidance deviation between environmental hazards and population movement trends.

[0099] Risk Resonance Submodule: The risk resonance intensity function is used to reveal the deep coupling mechanism between external mechanical disturbances and internal group disorder. The presence of resonance activates the dynamic disturbance propagation weight matrix.

[0100] Risk propagation and instability prediction module: Based on a dynamic disturbance propagation weight matrix, a heterogeneous personnel graph is constructed. An improved webpage ranking method is used to calculate the potential energy of risk influence propagation. Symbolic time series analysis and Markov analysis are performed on its historical disorder sequences. Relative entropy is used to measure the strength of behavioral inertia locking. The two methods are coupled to obtain a group-level instability prediction value. Specifically, it consists of the following sub-steps:

[0101] Heterogeneous Graph and Propagation Potential Submodule: Constructs dynamic heterogeneous graphs to simulate and calculate the propagation potential of risk impact.

[0102] Inertial Lockout Judgment Submodule: To quantify whether an individual's behavior has fallen into an inertial lockout state, symbolic time series analysis is performed on the historical multi-scale motion disorder index time series of individuals in the neighborhood of each node.

[0103] Instability prediction submodule: For each node marked as high risk, it integrates the potential energy of its risk influence propagation with the strength of behavioral inertia locking to generate a group-level instability prediction value.

[0104] Criticality Level Module: For individuals whose predicted values ​​exceed the limits, a comprehensive criticality level is output as the basis for graded early warning using motion trend prediction algorithms, spatial intersection detection algorithms, and shortest path algorithms. This is specifically divided into the following sub-steps:

[0105] Trajectory extrapolation submodule: Based on the current motion state of high-risk individuals, it generates the future sector-shaped spatial envelope region by introducing short-term trajectory extrapolation with directional deviation.

[0106] Spatial Intersection Submodule: Calculates the proportion of the spatial intersection volume between the trajectory envelope and the hazard source buffer zone in the building information model, and quantifies the possibility of it intruding into the high-risk area.

[0107] The crisis level generation submodule combines path complexity, risk propagation potential, and intersection degree to generate a graded comprehensive crisis level through a weighted formula to drive graded alarms.

[0108] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A construction crisis early warning method, characterized by, Comprise: S10, based on UWB base station collected high frequency three-dimensional coordinate data, through the sliding time window combined with wavelet packet energy entropy, direction autocorrelation attenuation ratio and speed change rate quantile, using principal component analysis fusion generated multi-scale motion disorder index; S20, according to the multi-scale motion disorder index to identify the sustained instability personnel, to its trajectory Bayesian change point segmentation, combined with hidden Markov model state transition spectrum radius and building information model extracted environmental parameters to build weighted graph, calculate the perceived load integral, generate cognitive mismatch degree index; S30, the cognitive mismatch degree is mapped to the space grid, the local Moran index is calculated to identify the risk cluster, the density gradient is fused with the mechanical operation sequence resonance result, and the dynamic interference propagation weight matrix is constructed to represent the spatial influence of external disturbance on risk diffusion; S40, based on the dynamic interference propagation weight matrix, the personnel heterogeneous graph is constructed, the improved web page ranking is used to calculate the risk influence force propagation potential, and the historical disorder sequence is symbolized time series analysis and Markov analysis, the relative entropy is used to measure the behavior inertia locking strength, and the two are coupled to obtain the group level instability prediction value; S50, for the individuals whose prediction value exceeds the limit, the motion trend prediction algorithm, space intersection detection algorithm and shortest path algorithm are used to output the comprehensive criticality grade as the grading early warning basis; S20 is specifically divided into the following substeps: For the personnel marked as persistent instability, the trajectory data is retrieved and the Bayesian change point detection algorithm is applied to identify the motion mode mutation point, and the trajectory is divided into multiple behavior segments; Each behavior segment is discretized into a spatial node to construct a path weighted graph, and a hidden Markov model is fitted to extract the behavior predictability, and an integrated environmental pressure index is constructed in combination with the environmental parameters analyzed by the building information model; The perceived load integral is calculated using the trajectory-environment coupling curvature integral function, and the cognitive mismatch degree is calculated according to the perceived load integral, which specifically includes: parameter initialization is performed according to the actual trajectory from the start point to the end point of the behavior segment, and the weight value is dynamically adjusted according to the semantic label of the building information model; numerical integration is performed on the trajectory-environment coupling curvature integral function along the trajectory to obtain the perceived load integral of the behavior segment; select the paragraph corresponding to the lowest predictability in all behavior segments, and take the highest perceived load total value as the basis to generate the cognitive mismatch degree according to the cognitive generation formula; The specific formula of the trajectory-environment coupling curvature integral function is: wherein, is a personnel trajectory parameterized curve, is a trajectory curvature, reflecting a direction mutation, is a decay constant, controlling the sensitivity of the net height change, is a net height change along the trajectory, is a lighting variation coefficient, is a ground friction coefficient gradient, is a weight coefficient; selecting the passage with the lowest predictability among all passages of the behavior segment is based on the formula generating a degree of cognitive mismatch, is the spectral radius of the state transition matrix of the behavior segment; S30 is specifically divided into the following substeps: The cognitive mismatch degree of the construction personnel is mapped to the space grid, and the local Moran index is calculated to identify the high-risk space cluster area; Based on the cosine of the angle between the crowd density gradient and the carbon monoxide concentration gradient, the crowd guiding distortion coefficient is constructed to quantify the guiding deviation of the environmental hazards and the personnel movement trend; the crowd guiding distortion coefficient is defined as The index reflects the deviation between the crowd movement trend and the harmful environment diffusion direction, is the crowd density gradient vector field; The risk resonance strength function is used to reveal the deep coupling mechanism between external mechanical disturbance and internal group disorder. If resonance exists, the active dynamic interference propagation weight matrix is activated. The specific formula is where T is the sliding time window, is the crowd density field, representing the number of people per unit area near the spatial position at time t, is the crowd density Laplace, whose negative value area corresponds to the density peak or aggregation vortex, which is a structural marker of group instability, is the time derivative of the heavy machinery start-stop signal extracted from the device Internet of Things log, which produces a pulse signal at the start-stop moment, representing the instantaneous impact of external disturbance; is the spectral coherence distance of the mechanical working cycle and the crowd movement main frequency, is the empirical scale parameter; if the value is greater than the set threshold, it is determined that there is significant resonance, and the active dynamic interference propagation weight matrix is activated S40 is specifically divided into the following substeps: A dynamic heterogeneous graph is constructed to simulate and calculate the risk influence propagation potential; the dynamic evolution equation of the heterogeneous graph risk propagation potential is used to accurately simulate the nonlinear propagation process of the risk on the dynamic heterogeneous graph, and the specific formula is wherein, represents the risk influence propagation potential of node i at time step k+1, represents the risk influence propagation potential of node j at time step k, is the damping factor of node i, and the formula is The value of the damping factor is adaptively adjusted according to the interference resonance strength of the environment in which the individual is located, a is a fixed constant, is a normalization coefficient, is the risk resonance strength, is the edge weight from neighbor j to i in the dynamic interference propagation weight matrix, and M is the total number of nodes, is the set of all neighbor nodes of node i, is the edge weight from node j to node i at time step k, is the edge weight from node j to node m at time step k, is a propagation efficiency function, is the information propagation delay from node j to node i at time step k, is a time interval; To quantify whether individual behavior is trapped in inertial lock state, the historical multi-scale motion disorder index time series of individuals in each node's neighborhood are analyzed as symbolic time series; the original time series are reduced dimensionally by piecewise aggregate approximation and mapped to 6 discrete symbols, realizing the abstract conversion from continuous dynamics to behavior state sequence, then based on the state transition frequency of all neighborhood individuals' symbolic sequences, Markov transition matrix M is constructed, and the stationary distribution of Markov chain is further solved , i.e. satisfying and long-term behavior distribution; on this basis, using relative entropy as the core measurement tool, the , is the 6-dimensional uniform distribution, representing the ideal behavior benchmark of complete randomness and irregularity, and the relative entropy quantifies the information deviation between the actual behavior distribution and the ideal random distribution: if the behavior is diverse and disordered, , ; if the behavior is highly concentrated in a few states, then , indicating that it has fallen into behavior inertial lock; For each node marked as high-risk, the risk influence propagation potential and the behavior inertia locking strength are fused to generate a group-level instability prediction value. The specific formula for generating the group-level instability prediction value is wherein, is the risk influence propagation potential, and Sigmoid function The propagation potential is normalized as the activation probability of the risk amplification capability, The rigidity degree of the neighborhood behavior pattern is depicted, and the product thereof reflects the synergistic danger effect of the high propagation individual in the low elasticity group environment.

2. The construction crisis early warning method of claim 1, wherein Based on the high frequency three-dimensional coordinate data collected by the UWB base station, the multi-scale motion disorder index is generated by combining the wavelet packet energy entropy, the direction autocorrelation attenuation ratio and the speed change rate quantile through the sliding time window, and the principal component analysis is used for fusion; Three-dimensional coordinate data is collected by UWB positioning label to form trajectory data, wavelet packet decomposition is performed on the vertical acceleration based on the trajectory data through sliding window, and energy distribution entropy is calculated to quantify the vertical motion chaos; The horizontal direction change characteristics are analyzed, the vertical fluctuation entropy is standardized, the multi-dimensional motion characteristics are fused through principal component analysis, and the instantaneous disorder component is extracted; The HURST index is calculated by sliding the instantaneous turbulence component sequence, and if it remains above the threshold value, it is determined to be persistent instability, and the multi-scale motion turbulence index is output and the deep assessment is triggered.

3. The construction crisis early warning method of claim 1, wherein For individuals whose pre-judgment value exceeds the limit, the comprehensive criticality level is output as the basis for graded early warning through the motion trend prediction algorithm, spatial intersection detection algorithm and shortest path algorithm, which is divided into the following sub-steps: Based on the current motion state of high-risk individuals, the short-time trajectory extrapolation by introducing directional deviation is used to generate the fan-shaped space envelope region occupied in the future; The trajectory envelope and the spatial intersection volume ratio of the buffer zone of the dangerous source in the building information model are calculated to quantify the possibility of invading the high-risk area; Combined with path complexity, risk propagation potential and intersection degree, the weighted formula is used to generate a graded comprehensive criticality level to drive graded alarm.

4. A construction crisis warning system which executes a construction crisis warning method according to any one of claims 1 to 3, characterized by It includes: The acquisition and generation module: based on the high-frequency three-dimensional coordinate data collected by the UWB base station, the multi-scale motion turbulence index is generated by sliding time window combined with wavelet packet energy entropy, directional autocorrelation decay ratio and velocity change rate quantile, and principal component analysis is used for fusion; The identification and calculation module: according to the multi-scale motion turbulence index, the personnel with persistent instability are identified, the trajectory is segmented by Bayesian change point, the weighted graph is constructed combined with the state transition spectrum radius of hidden Markov model and the environmental parameters extracted from building information model, the perceptual load integral is calculated, and the cognitive mismatch degree index is generated; The mapping and construction module: the cognitive mismatch degree is mapped to the space grid, the local Moran index is calculated to identify risk clusters, the density gradient and mechanical operation sequence resonance results are fused to construct a dynamic interference propagation weight matrix to represent the spatial influence of external disturbance on risk diffusion; The risk propagation and instability prediction module: based on the dynamic interference propagation weight matrix, a heterogeneous personnel graph is constructed, the risk influence propagation potential is calculated by improved web page ranking, and symbolic time series analysis and Markov analysis are performed on its historical turbulence sequence, the behavior inertia locking strength is measured by relative entropy, and the two are coupled to get the group-level instability prediction value; The criticality level module: for individuals whose pre-judgment value exceeds the limit, the comprehensive criticality level is output as the basis for graded early warning through the motion trend prediction algorithm, spatial intersection detection algorithm and shortest path algorithm.

Images