Construction high-altitude falling accident intelligent monitoring and early warning method and system

Abstract

This invention relates to the field of construction fall monitoring technology, specifically to an intelligent monitoring and early warning method and system for construction high-altitude fall accidents. It introduces time-step-based behavioral pattern branches, dividing the regional distribution probability of different behavioral pattern branches; these branches include high-altitude operation mode and edge movement mode; a fusion control vector is constructed based on the scene risk vector and individual deviation vector, and weight vectors and activation masks are calculated to calculate the output channel vector; a fusion feature column vector is established, and the fall risk probability is predicted through a fully connected layer to determine whether an early warning is triggered. If triggered: the fall point is calculated based on free fall, the rope length is calculated in conjunction with the safety rope anchor point, and it is determined whether to calculate the impact velocity and impact force. The elasticity and damping of the safety rope are introduced to calculate the applied force, and a system of first-order ordinary differential equations is established; displacement and impact velocity conditions are set, and the final fall point is calculated in conjunction with the anchor point vector, displacement, and the worker's rope-direction displacement.

Description

Technical Field

[0001] This invention relates to the field of construction fall monitoring technology, specifically to an intelligent monitoring and early warning method and system for construction high-altitude fall accidents. Background Technology

[0002] Traditional intelligent monitoring for construction fall accidents relies on rigid area coding (such as dividing the work area into zones A / B / C) and using fixed thresholds to determine whether sensor data (such as position and acceleration) exceeds limits. However, high-altitude work activities are highly dynamic and ambiguous: the same area (such as zone A) may simultaneously contain resting personnel and high-risk workers, which cannot be distinguished by a single area label; when personnel move along the boundaries, the geographic location coding becomes ineffective; and the temporal changes in safety rope tension, attitude angle, and other characteristics are not deeply analyzed, resulting in a high false alarm rate and a significant risk of missed alarms.

[0003] In the accident prediction stage, existing technologies mostly use simplified free fall models, ignoring the mechanical buffering mechanisms of safety ropes (such as elastic deformation and impact fracture threshold), and cannot distinguish between scenarios such as effective braking of safety ropes, rope breakage, and buffering oscillations. This results in large deviations in fall prediction and delays in rescue deployment for workers and construction areas. Summary of the Invention

[0004] In view of the above-mentioned shortcomings of the existing technology, the present invention provides an intelligent monitoring and early warning method and system for construction high-altitude fall accidents, which can effectively solve the problem that the existing technology is difficult to combine with actual construction scenarios, resulting in poor accuracy of fall accident monitoring and early warning during the construction process.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] This invention provides an intelligent monitoring and early warning method for construction fall accidents, comprising at least:

[0007] Collect observation vectors from workers within the construction area and establish a single-moment state vector;

[0008] Fall risk prediction using time-series depth models includes:

[0009] The multi-channel signal matrix is ​​obtained based on the state vector at a single moment, and the context vector is output.

[0010] Introduce time-step-based behavioral pattern branches and divide the regional distribution probability of different behavioral pattern branches;

[0011] Behavioral pattern branches include high-altitude operation mode and edge movement mode;

[0012] A fusion control vector is constructed based on the scenario risk vector and the individual deviation vector, and the weight vector and activation mask are calculated to calculate the output channel vector.

[0013] Establish a fused feature column vector, predict the probability of fall risk through a fully connected layer, and determine whether to trigger an early warning. If triggered:

[0014] Calculate the point of impact based on free fall, calculate the rope length based on the safety rope anchor point, and determine whether to calculate the impact velocity and impact force.

[0015] The elasticity and damping of the safety rope are introduced to calculate the applied force, and a system of first-order ordinary differential equations is established.

[0016] By setting displacement and impact velocity conditions, and combining the anchor point vector, displacement, and rope direction displacement of the worker, the final fall point is calculated.

[0017] The method for outputting the context vector is as follows:

[0018] Set a sliding time window;

[0019] Construct a timing matrix and obtain a multi-channel signal matrix through transpose operation;

[0020] Extract a local sequence from the input multi-channel signal matrix as the context segment for that time step;

[0021] Perform one-dimensional convolutional layer processing and output the context vector through the ReLU activation function.

[0022] The method for determining the probability of regional distribution is as follows:

[0023] For each time step, construct the regional distribution probability of the behavior at that time step across all pattern branches;

[0024] After inputting the context vector into a two-layer perceptron MLP, it is mapped to a real number vector to obtain the regional distribution probability of the time walk belonging to different behavioral pattern branches.

[0025] The scenario risk vector is constructed using temperature, humidity, and wind speed; the individual deviation vector is constructed using fatigue and focus.

[0026] Furthermore, the specific steps for calculating the output channel vector are as follows:

[0027] Obtain all channel dimensions, extracted channel features, activation mask, fusion weights of behavior pattern branches, and calculate the probability of behavior pattern branches to output channel vectors.

[0028] If an early warning is triggered at a certain time step, the trigger time is determined, the difference between the trigger time and the current time is calculated, and the free fall model is introduced to calculate the fall point.

[0029] Furthermore, the method for calculating the impact velocity and impact force is as follows:

[0030] Set the mass of the worker, obtain the velocity vector of the worker, and calculate the impact velocity when the safety rope is fully stretched by combining the gravitational acceleration and the difference.

[0031] The impact force is calculated based on the impact velocity, mass, and buffer time.

[0032] Furthermore, the method for determining the applied force is as follows:

[0033] Calculate the difference between the current rope length and the rope length when it is stationary and not stretched, calculate the rate of change of rope length, and input the elastic constant and damping coefficient of the safety rope to calculate the applied force.

[0034] A smart monitoring and early warning system for construction falls from heights, applied to the aforementioned smart monitoring and early warning method for construction falls from heights, includes:

[0035] The state vector establishment module collects the observation vectors of workers in the construction area and establishes a state vector at a single moment.

[0036] The fall risk prediction module obtains a multi-channel signal matrix based on the state vector at a single moment and outputs a context vector.

[0037] Introduce time-step-based behavioral pattern branches and divide the regional distribution probability of different behavioral pattern branches;

[0038] Behavioral pattern branches include high-altitude operation mode and edge movement mode;

[0039] A fusion control vector is constructed based on the scenario risk vector and the individual deviation vector, and the weight vector and activation mask are calculated to calculate the output channel vector.

[0040] A fused feature column vector is established, and the probability of fall risk is predicted through a fully connected layer;

[0041] The crash site prediction module determines whether to trigger an alert based on the probability of a fall risk. If triggered:

[0042] Calculate the point of impact based on free fall, calculate the rope length based on the safety rope anchor point, and determine whether to calculate the impact velocity and impact force.

[0043] The elasticity and damping of the safety rope are introduced to calculate the applied force, and a system of first-order ordinary differential equations is established.

[0044] By setting displacement and impact velocity conditions, and combining the anchor point vector, displacement, and rope direction displacement of the worker, the final fall point is calculated.

[0045] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0046] The technical solution provided by this invention has the following advantages compared with the known prior art:

[0047] By calculating the probability distribution of workers belonging to different behavioral patterns at each time point, the ambiguity of behavior is quantified; an independent feature extraction channel is designed for each behavioral pattern, and differentiated one-dimensional convolutional kernels are used to capture the key features of specific patterns; finally, weights and masks are dynamically generated by combining environmental parameters and individual states, automatically activating relevant branches and suppressing irrelevant noise, thereby enabling fine-grained perception and monitoring of construction behavior.

[0048] By calculating the distance between the fall point and the safety rope anchor point, if it exceeds the rope length threshold, an impact force determination is triggered. Based on the personnel mass, buffer time, and impact velocity, the instantaneous tension borne by the safety rope is calculated to determine whether to model the safety rope as a spring-damped system. This simulates the reciprocating oscillation process of the personnel caused by the rope's elasticity and uses Newton's laws to calculate the motion trajectory in real time until the velocity or acceleration is lower than the rest threshold. Finally, the precise fall point after the buffering is completed is output, providing crucial information for the deployment of rescue buffer pads and personnel evacuation. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0050] Figure 1 This is a schematic diagram of the overall method of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0052] The present invention will be further described below with reference to embodiments.

[0053] Example 1 (see Figure 1 A method for intelligent monitoring and early warning of construction fall accidents, comprising at least:

[0054] Collect sensor observation vectors from workers within the construction area, including:

[0055] Vertical acceleration (Reflects the vertical acceleration state of the worker's body or equipment), rate of change of acceleration (current vertical acceleration minus the vertical acceleration at the previous sampling time), and the coordinate position of the worker. (Position of the worker), velocity vector of the worker Attitude angle (Including three-dimensional attitude angle information: roll angle, pitch angle, yaw angle, used to determine whether personnel are tilting, rolling, or unbalanced, etc.), attitude angle change rate Safety rope tension (Current stress on the safety rope), Rate of change of stress on the safety rope , This indicates the area coding, such as the operating platform, scaffold edge, etc., and the entire high-altitude operation area is rigidly divided into several discrete, fixed area blocks (such as area A, area B, area C, etc.).

[0056] Then, the collected data of various types are standardized and scaled to form a single-time state vector. :

[0057]

[0058] Indicates the workers, Indicates the time of data collection;

[0059] This allows for the fusion of multiple data sets, enabling subsequent assessments of whether workers are at risk of falling.

[0060] Then, risk prediction is performed using a time-series deep model, resulting in:

[0061] Set sliding time window For example (e.g., 2 seconds, 100 frames);

[0062] Build length is A time series matrix with a single-time state vector dimension of D. Considering the complexity of construction behavior and the heterogeneity of behavior within the window, unified modeling may fail. Therefore, the time series matrix is ​​transposed to obtain a multi-channel signal matrix. ;

[0063] In risk monitoring of workers at height, relying solely on single-point data (such as current location or sensor readings) may not accurately determine the existence of a fall risk, as falls often exhibit dynamic characteristics. Therefore, to avoid short-term fluctuations interfering with the assessment:

[0064] From the input multi-channel signal matrix China and Israel Centered on the time step, extract a segment of length [time step]. The local sequence, as a context fragment for that time step:

[0065] ,in, , This represents the half width of the sliding window, in the formula. The index is a slice notation for tensor / matrix indexes, describing the method used to retrieve all elements. Indicates from arrive The elements, the whole Represented as taking a multi-channel signal matrix All lines, and arrive The elements (column elements);

[0066] Then, a one-dimensional convolutional layer is performed. Process the data and output the context vector using the ReLU activation function. :

[0067] Extract the state changes of that time step within the context to capture behavioral trends and environmental changes.

[0068] Traditional methods categorize work scenarios solely through rigid regional coding, failing to reflect behavioral diversity and boundary ambiguity (they don't consider changes in personnel behavior, only using geographical location for regional division; they cannot handle ambiguous behaviors of personnel moving across boundaries or regions). Therefore, a time-step-based behavioral pattern branching approach is introduced, dividing the regional distribution probabilities of different behavioral pattern branches, including:

[0069] For each time step (Sliding time window) (within), construct the regional distribution probability of the behavior at this time step across all pattern branches. , , Indicates at time step Below, the workers belong to the first category. The probability of each being a branch of a behavioral pattern;

[0070] Context vector The input to a two-layer perceptron MLP is mapped to a real number vector. The probability distribution of walking at that time belonging to different behavioral pattern branches is obtained. .

[0071] Because the behavioral patterns of workers at heights vary significantly in different scenarios (such as near platform edges or at high altitudes), it is not advisable to use a single model for uniform feature extraction. Forcing a uniform model would mask subtle but crucial regional characteristics and reduce the accuracy of subsequent risk identification.

[0072] For each operational area behavior mode branch (e.g., edge movement mode, high-altitude operation mode, etc.), a separate feature extraction branch is set up, and each feature extraction branch uses a separate one-dimensional convolutional kernel. (Including the number of input channels, the number of output channels, and the kernel width, the edge movement mode corresponds to 3, 16, and 5 respectively; the high-altitude operation mode corresponds to 5, 32, and 7 respectively), extracting features under this behavior mode branch (edge ​​movement mode: lateral drift of position coordinates, horizontal component of velocity vector, periodic fluctuation of safety rope force, thereby identifying slippage or missteps when walking along the edge; high-altitude operation mode: vertical acceleration, pitch angle and pitch angle change amplitude, rate of change of safety rope force, peak value of safety rope tension, thereby detecting continuous rollover caused by high-altitude instability), that is, using corresponding convolution kernels for different construction blocks to capture regional differences:

[0073]

[0074] in, , Indicates the first There are 1 behavioral pattern branch, with a total of 1 Since different behavioral patterns may exist at the current time point, multiple feature extraction branches are used in parallel to perform multi-angle extraction, thereby maintaining pattern differentiation and ensuring that the sensitivity to high-risk action features is not averaged out.

[0075] In actual high-altitude operations, not all behavioral patterns corresponding to all areas (such as different construction locations or risk levels) are applicable to every worker or every work scenario. For example:

[0076] Some workers may be in peripheral areas, but their behavior is more similar to that of those in the central areas.

[0077] Therefore, statically using all behavior pattern branches is not only a waste of computing resources, but may also lead to misjudgment or omission. In other words, not all behavior pattern branches are reliable under the current operators and scenarios. It is necessary to dynamically select which branches to participate in the fusion and assign different weights.

[0078] To achieve dynamic selection and personalized adaptation of behavioral pattern branches in different areas, a fusion control vector is constructed based on the scenario risk vector (reflecting the global state of the current working environment, such as temperature, humidity, and wind speed) and the individual deviation vector (reflecting the current worker's own characteristics, such as fatigue or concentration; both are in the range of 0-1 and can be identified by acquiring the worker's physiological data and using a neural network model, or by mapping various physiological data to the same scale for weighted calculation). Adjust which behavioral pattern branches should be activated and their weights, specifically:

[0079] Using a two-layer MLP network The mapping is a weight vector (controlling the proportion of each branch in the final fusion) and an activation mask for each region's behavioral pattern branch, including:

[0080] The first layer extracts the intermediate representation. ;

[0081]

[0082] in, , This represents the weight matrix and bias vector of the first layer;

[0083] The second layer generates two outputs: a weight vector and an activation mask.

[0084]

[0085] Then, the fusion weights for all regions Softmax normalization is performed to convert the scores of all regions into probabilities. In this way, at each time step, a worker can belong to multiple behavior pattern branches at the same time. Each behavior pattern branch corresponds to a probability weight. Therefore, at each time point, the behavior of a single worker can belong to multiple behavior pattern branches at the same time, and each branch has a corresponding probability weight.

[0086] Then, Gumbel-Softmax sampling (converting the continuous activation signal into a discrete binary decision) is used to evaluate the activation mask candidate values. Approximate binarization is performed (on top of the unprocessed activation function) to obtain an activation mask that produces 0 / 1. This prevents irrelevant or unreliable branches from participating in the decision-making process, reducing noise and computational burden.

[0087] For the Layer, time step The fusion formula is:

[0088]

[0089] in, Indicates the first Layer time step The output channel vector, This is used here to retrieve all channel dimensions (channel 1 to channel 2). (each channel), is a slice notation for a tensor / matrix, used to extract the representation of all channels at that position. Indicates the first The behavioral pattern branch in the first Layer time step Extracted channel features, Indicates the first The activation mask for each behavioral pattern branch is used to determine whether the branch is active; it is either 1 or 0. Indicates the first The fusion weight of each behavioral pattern branch.

[0090] Then you can take it. fused feature column vector ;

[0091] A fully connected layer (with trained weight matrices and bias vectors, including a sigmoid activation function for outputting probability values, all common configurations) is established to map the fused feature column vectors to risk probabilities, thereby obtaining the first... The predicted fall risk probability for each time step, relative to the aforementioned sliding time window. Repeat the steps of the fully connected layer to obtain the fall risk probability at all time steps and establish a risk probability sequence. If there exists a time step... If the probability of a fall is greater than or equal to the threshold, an early warning mechanism is triggered (voice warning broadcast to the operator's terminal and warning from the control center).

[0092] Furthermore, if there exists a future time step If the warning mechanism is triggered, the time when the warning mechanism is triggered is determined, and the difference between the trigger time and the current time is calculated. And by establishing a fall risk accident point model to predict the fall accident points of workers, we have:

[0093] The theoretical fall point is calculated based on the free fall model (without a safety rope, with the safety rope broken, or not being worn):

[0094]

[0095] in, This represents the predicted future fall point for the worker (a three-dimensional vector). , All are three-dimensional vectors. Representing gravitational acceleration (a three-dimensional vector), based on a physical model of free fall motion, and combining current velocity and position, it predicts the location of a worker's fall accident at a future moment.

[0096] Furthermore, obtain the coordinates of the anchor point at the other end of the safety rope worn by the worker. (Three-dimensional vector), typically installed on a sturdy, reliable structure (such as building beams, platform edges, or dedicated anchor points), to ensure the safety rope effectively supports the worker, taking into account whether the worker will trigger the maximum effective rope length threshold during the fall:

[0097] Calculate the safety rope length:

[0098]

[0099] in, Indicates the rope length and describes the distance between the fall point and the anchor point.

[0100] like If the rope length is less than or equal to the threshold, it means the safety rope is not fully extended, and the worker will not continue to fall. This is the actual point of impact;

[0101] like If the length exceeds the rope length threshold, it means the safety rope is fully stretched, proceed to the next step:

[0102] Considering the impact when the safety rope is fully stretched during free fall, whether the safety rope will exceed its bearing capacity, we have:

[0103] Calculate the rope tension to determine if it will break:

[0104] Let the quality of the workers be... The impact velocity at the moment the rope is fully pulled is :

[0105]

[0106] Impact The estimate is:

[0107]

[0108] in, The buffering time describes the time from when a worker touches the safety rope in free fall until the safety rope is fully stressed, and is usually set to 0.2-0.5s.

[0109] If the impact force If the impact force exceeds the threshold, the safety rope is determined to break, and the worker's accidental fall point is determined to be... .

[0110] If the impact force If the impact force is less than or equal to the impact force threshold, the safety rope is determined not to break, and the process proceeds to the next elastic buffer modeling step.

[0111] When the safety rope is fully extended, it does not come to an instantaneous stop. Instead, due to the rope's elasticity and damping, the worker experiences a buffering process similar to spring vibration—first being pulled back by the rope, then bouncing back up, and gradually stabilizing. In this embodiment, the safety rope is considered as a one-dimensional spring-damped system, and the applied force is:

[0112]

[0113] in, This indicates the force exerted by the safety rope on the worker. This represents the elastic constant (stiffness) of the safety rope. This represents the rope length at the current moment and the rope length when it is at rest and not stretched. The difference between them Indicates the damping coefficient. Indicates the rate of change of rope length .

[0114] Furthermore, the equation of motion of the worker's center of mass along the rope length satisfies Newton's second law:

[0115]

[0116] in, This indicates the acceleration of the worker in the direction of the rope.

[0117] Transform into a system of first-order ordinary differential equations:

[0118]

[0119] in, Indicates the speed of the workers;

[0120] Set initial conditions,

[0121] The simulation integration process continues until one of the following conditions is met:

[0122] If the acceleration or speed of a person at the current location is less than the stationary threshold, then the movement stops.

[0123] When the current time point in the simulation reaches the maximum simulation time, a maximum simulation time limit is set to prevent the simulation from running indefinitely. Once this value is reached, the simulation process is forcibly terminated to avoid the problem of calculating the actual position of the worker from the start of the fall to the final stationary position after all the buffering processes.

[0124] Therefore, the final point of impact for the worker was:

[0125]

[0126] in, This indicates the final fall point of the worker after the simulated buffering process. This represents the unit vector pointing from the anchor point to the point of impact. This indicates the displacement of the worker along the rope direction when the simulation ends.

[0127] This enables the monitoring and early warning of worker fall accidents in the future, allowing the control center to implement corresponding safety risk management measures, including but not limited to evacuating construction workers from the fall site and quickly installing rescue buffer equipment at the fall site, thereby improving the ability to protect the personal safety of workers.

[0128] A smart monitoring and early warning system for construction falls from heights, applied to the aforementioned smart monitoring and early warning method for construction falls from heights, includes:

[0129] The state vector establishment module collects the observation vectors of workers in the construction area and establishes a state vector at a single moment.

[0130] The fall risk prediction module obtains a multi-channel signal matrix based on the state vector at a single moment and outputs a context vector.

[0131] Introduce time-step-based behavioral pattern branches and divide the regional distribution probability of different behavioral pattern branches;

[0132] Behavioral pattern branches include high-altitude operation mode and edge movement mode;

[0133] A fusion control vector is constructed based on the scenario risk vector and the individual deviation vector, and the weight vector and activation mask are calculated to calculate the output channel vector.

[0134] A fused feature column vector is established, and the probability of fall risk is predicted through a fully connected layer;

[0135] The crash site prediction module determines whether to trigger an alert based on the probability of a fall risk. If triggered:

[0136] Calculate the point of impact based on free fall, calculate the rope length based on the safety rope anchor point, and determine whether to calculate the impact velocity and impact force.

[0137] The elasticity and damping of the safety rope are introduced to calculate the applied force, and a system of first-order ordinary differential equations is established.

[0138] By setting displacement and impact velocity conditions, and combining the anchor point vector, displacement, and rope direction displacement of the worker, the final fall point is calculated.

[0139] A computer device includes a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the method.

[0140] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method.

[0141] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligent monitoring and early warning of construction fall accidents, characterized in that, include: Collect observation vectors of workers in the construction area, including vertical acceleration, rate of change of acceleration, coordinate position of personnel, velocity vector of personnel, attitude angle, rate of change of attitude angle, force on safety rope, rate of change of force on safety rope, and area code, and establish a state vector at a single moment. Fall risk prediction using time-series depth models includes: The multi-channel signal matrix is ​​obtained based on the state vector at a single time moment, and the context vector is output. Introduce time-step-based behavioral pattern branches and divide the regional distribution probability of different behavioral pattern branches; For each time step, construct the regional distribution probability of the behavior at that time step across all pattern branches; After inputting the context vector into the two-layer perceptron MLP, it is mapped to a real number vector to obtain the regional distribution probability of the time walk belonging to different behavioral pattern branches. Behavioral pattern branches include high-altitude operation mode and edge movement mode; A fusion control vector is constructed based on the scenario risk vector and the individual deviation vector, and the weight vector and activation mask are calculated to calculate the output channel vector. Establish a fused feature column vector, predict the probability of fall risk through a fully connected layer, and determine whether to trigger an early warning. If triggered: Calculate the point of impact based on free fall, calculate the rope length based on the safety rope anchor point, and determine whether to calculate the impact velocity and impact force. The elasticity and damping of the safety rope are introduced to calculate the applied force, and a system of first-order ordinary differential equations is established. By setting displacement and impact velocity conditions, and combining the anchor point vector, displacement, and rope direction displacement of the worker, the final fall point is calculated.

2. The intelligent monitoring and early warning method for construction high-altitude fall accidents according to claim 1, characterized in that, The method for outputting the context vector is as follows: Set a sliding time window; Construct a timing matrix and obtain a multi-channel signal matrix through transpose operation; Extract a local sequence from the input multi-channel signal matrix as the context segment for that time step; Perform one-dimensional convolutional layer processing and output the context vector through the ReLU activation function.

3. The intelligent monitoring and early warning method for construction high-altitude fall accidents according to claim 1, characterized in that, The scenario risk vector is constructed using temperature, humidity, and wind speed. Individual deviation vectors are constructed using fatigue and focus.

4. The intelligent monitoring and early warning method for construction high-altitude fall accidents according to claim 1, characterized in that, The specific calculation of the output channel vector is as follows: Obtain all channel dimensions, extracted channel features, activation mask, fusion weights of behavior pattern branches, and calculate the probability of behavior pattern branches to output channel vectors.

5. The intelligent monitoring and early warning method for construction high-altitude fall accidents according to claim 1, characterized in that, If an early warning is triggered at a certain time step, the trigger time is determined, the difference between the trigger time and the current time is calculated, and the free fall model is introduced to calculate the fall point.

6. The intelligent monitoring and early warning method for construction high-altitude fall accidents according to claim 5, characterized in that, The method for calculating the impact velocity and impact force is as follows: Set the mass of the worker, obtain the velocity vector of the worker, and calculate the impact velocity when the safety rope is fully stretched by combining the gravitational acceleration and the difference. The impact force is calculated based on the impact velocity, mass, and buffer time.

7. The intelligent monitoring and early warning method for construction high-altitude fall accidents according to claim 1, characterized in that, The method for determining the applied force is as follows: Calculate the difference between the current rope length and the rope length when it is stationary and not stretched, calculate the rate of change of rope length, and input the elastic constant and damping coefficient of the safety rope to calculate the applied force.

8. An intelligent monitoring and early warning system for construction high-altitude fall accidents, applied to the intelligent monitoring and early warning method for construction high-altitude fall accidents as described in any one of claims 1-7, characterized in that, include: The state vector establishment module collects the observation vectors of workers in the construction area and establishes a state vector at a single moment. The fall risk prediction module obtains a multi-channel signal matrix based on the state vector at a single moment and outputs a context vector. Introduce time-step-based behavioral pattern branches and divide the regional distribution probability of different behavioral pattern branches; Behavioral pattern branches include high-altitude operation mode and edge movement mode; A fusion control vector is constructed based on the scenario risk vector and the individual deviation vector, and the weight vector and activation mask are calculated to calculate the output channel vector. A fused feature column vector is established, and the probability of fall risk is predicted through a fully connected layer; The crash site prediction module determines whether to trigger an alert based on the probability of a fall risk. If triggered: Calculate the point of impact based on free fall, calculate the rope length based on the safety rope anchor point, and determine whether to calculate the impact velocity and impact force. The elasticity and damping of the safety rope are introduced to calculate the applied force, and a system of first-order ordinary differential equations is established. By setting displacement and impact velocity conditions, and combining the anchor point vector, displacement, and rope direction displacement of the worker, the final fall point is calculated.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images