Iot-based power aerial work safety monitoring system and method

Abstract

The application relates to the technical field of Internet of Things monitoring, in particular to a power high-altitude operation safety monitoring system and method based on the Internet of Things, which comprises a state sensing module, a time sequence analysis module, a gradient determination module, a feature fusion module and a state evaluation module. In the application, a multi-source sensing matrix is generated by matching the tension value and the vertical acceleration according to time, deviation interference caused by single collection is eliminated, a sudden change sequence is screened by combining difference extraction and window rate comparison to identify a stress mutation point, the beginning of an operation abnormality is locked to break through the lag bottleneck of manual monitoring, an abnormal sequence is constructed by counting the difference value of a synchronous node and calculating the extreme value amplitude ratio, microcosmic dangerous omens in normal signals are mined, a feature vector is established by using time regularization to align abnormal waveforms and obtain a mutation interval, a time lag relationship in an instability evolution process is quantified, an early warning instruction is issued under an abnormal risk state according to a classification model, and an active risk interception defense line is constructed.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) monitoring technology, and in particular to an IoT-based safety monitoring system and method for high-altitude power operations. Background Technology

[0002] The field of IoT monitoring technology encompasses a comprehensive system of technologies for real-time sensing, status acquisition, and remote monitoring of various target objects in the physical world through sensors, communication networks, and data acquisition devices. Based on the deployment of sensor nodes, this field connects geographically dispersed acquisition terminals to a backend platform using wireless or wired communication protocols, enabling the continuous acquisition and uploading of multi-dimensional physical quantities such as temperature, pressure, displacement, electrical parameters, and ambient gas concentration. In the power industry, IoT monitoring technology is widely used for transmission line status sensing, substation equipment operation monitoring, and safety management of personnel at power construction sites. By deploying various sensors at key locations, field data is aggregated to a monitoring center. Combined with data comparison and rule-based judgment mechanisms, abnormal states are identified and alerted, thereby assisting maintenance personnel in understanding the dynamics of the site.

[0003] Traditional power high-altitude operation safety monitoring systems refer to systems that monitor and control the personal safety status of workers and the hazardous factors in the working environment in real time for high-altitude operation scenarios in the power industry, such as pole climbing, line construction, and substation high-altitude maintenance. These systems address safety accidents caused by fall risks, violations of operating procedures, and sudden changes in the on-site environment during high-altitude operations. Traditional systems collect human posture data by installing accelerometers and tilt sensors on workers' helmets, safety belts, or work clothes. They use altitude and barometric pressure sensors or GPS modules to obtain the personnel's location and altitude information. The collected data is transmitted to a ground monitoring terminal via a wireless communication module. Simultaneously, cameras are deployed in the work area to obtain on-site video images. Monitoring personnel manually verify and judge the workers' helmet wearing status, safety belt engagement status, and boundary violations by combining sensor values ​​and video footage. An audible and visual alarm device alerts on-site personnel when an anomaly is detected.

[0004] Traditional technologies rely on manual observation of monitoring videos combined with data collected by sensors to determine the status. This operating mode is prone to causing personnel fatigue and distraction in scenarios with continuous input of massive amounts of data. Subjective judgment inevitably suffers from cognitive delays and differences in standards, making it difficult to quickly capture subtle changes in force and abnormal posture characteristics in the instant of sudden instability. The delayed response time often misses the best time for intervention. The monotonous fixed threshold comparison cannot truly reproduce the dynamic force evolution process of workers in complex high-altitude environments, which can easily lead to serious false alarms or missed alarms. Summary of the Invention

[0005] To address the technical problems of traditional technologies that rely on manual observation of monitoring videos combined with sensor-based data collection for status judgment, which are prone to causing personnel fatigue and distraction under continuous input of massive data, subjective judgment suffers from unavoidable cognitive delays and standard differences, making it difficult to quickly capture subtle changes in force and abnormal posture characteristics in the instant of sudden instability, and the delayed response time often misses the best intervention opportunity, and the monotonous fixed threshold comparison cannot truly reproduce the dynamic force evolution process of workers in complex high-altitude environments, which is prone to serious false alarms or missed alarms, this invention provides an Internet of Things-based power high-altitude operation safety monitoring system and method.

[0006] On the one hand, an IoT-based safety monitoring system for high-altitude power operations is provided, which includes: The state perception module acquires the seat belt tension and torso vertical acceleration through the IoT gateway, performs time sequence alignment matching based on the time record, and constructs a multi-source perception time sequence matrix. The time series analysis module extracts the transition time of the tension difference from positive to negative based on the multi-source sensing time series matrix, calculates the front window rate and the back window rate, filters sample points with the front window rate greater than a preset positive threshold and the back window rate less than a preset negative threshold, and generates a rate mutation sequence. The gradient determination module extracts the synchronization node difference based on the rate mutation sequence, calculates the proportion of sign-consistent difference, calculates the extreme value amplitude ratio for sampling points below a preset ratio threshold, extracts the moment when the extreme value amplitude ratio exceeds a preset upper limit threshold, and constructs a force anomaly sequence. The feature fusion module uses a dynamic time warping algorithm to align the force anomaly sequence with the vertical acceleration of the torso to obtain acceleration abrupt change points, calculates the time interval between the acceleration abrupt change points and the anomaly trigger time, and generates a critical instability feature vector. The state assessment module calls the critical instability feature vector, performs classification operations through a support vector machine model to extract abnormal risk states, and generates safety warning instructions.

[0007] As a further aspect of the present invention, the multi-source sensing time series matrix includes timestamps, spatial coordinates, and signal-to-noise ratio; the rate mutation sequence includes peak acceleration, rate of change, and relaxation time; the force anomaly sequence includes the action period, stress amplitude, and off-center load coefficient; the critical instability feature vector includes time delay, phase difference, and cross-correlation coefficient; and the safety warning instruction includes interruption priority, linkage control signal, and fault code.

[0008] As a further aspect of the present invention, the state sensing module includes: The data receiving submodule obtains the seat belt tension and torso vertical acceleration from the IoT gateway, configures the acquisition cycle for the IoT gateway, calibrates the time record of the seat belt tension and torso vertical acceleration according to the acquisition cycle, and extracts the timestamp to obtain the node synchronization time. The timing alignment submodule calls the node synchronization time amount, calculates the difference between the seat belt tension acquisition value associated time record and the torso vertical acceleration associated time record, and determines that when the difference is less than the alignment deviation benchmark value, it filters the value items in the corresponding time node and integrates them to generate a state alignment data group. The matrix construction submodule, based on the state-aligned data group, sets the seat belt tension acquisition values ​​as a horizontal element sequence and the torso vertical acceleration as a vertical element sequence. The horizontal and vertical element sequences are stored in a two-dimensional grid structure according to the same time node to establish a multi-source sensing time sequence matrix.

[0009] As a further aspect of the present invention, the time series analysis module includes: The transition time extraction submodule extracts tension sampling values ​​based on the multi-source sensing time series matrix, matches time series nodes, calculates the difference between tension sampling values ​​of adjacent nodes, determines the location where the sign of the difference changes from positive to negative, extracts the location-related time record, and generates the transition time when the tension difference changes from positive to negative. The window rate calculation submodule calls the moment when the tension difference changes from positive to negative, and uses the moment when the tension difference changes from positive to negative to extract a fixed time period on both sides of the time axis to define the pre-calculation window and the post-calculation window. Based on the tension change within the window divided by the time span, the pre-window rate and the post-window rate are calculated to obtain the window rate group. The mutation sequence screening submodule, based on the rate evolution trend of the window rate group, determines that the preceding window rate is greater than a preset positive threshold and the following window rate is less than a preset negative threshold, filters test sample points associated with window rate groups that conform to the evolution trend, merges test sample points according to time sequence, and establishes a rate mutation sequence.

[0010] As a further aspect of the present invention, the gradient determination module includes: The difference percentage statistics submodule extracts the synchronization node difference based on the rate mutation sequence, performs sign consistency check on the synchronization node difference, calculates the proportion of the number of sign-consistent differences to the total number of all records, and generates the sign-consistent difference percentage. The amplitude ratio calculation submodule compares the sign consistency difference ratio with a preset ratio threshold. For time-series sampling points whose values ​​are lower than the preset ratio threshold, it extracts the local tension extreme value and the normal operating amplitude. The local tension extreme value is divided by the normal operating amplitude to obtain the extreme value amplitude ratio. The abnormal sequence construction submodule extracts the target time nodes corresponding to the extreme value amplitude ratios that exceed the preset upper limit threshold, and performs merging and arrangement of the target time nodes according to the time sequence to construct the stress anomaly sequence.

[0011] As a further aspect of the present invention, the preset proportion threshold is determined by extracting the proportion of consistency differences of multiple sets of reference symbols within a historical normal operating period and calculating the arithmetic mean of the proportions of consistency differences of multiple sets of reference symbols. The preset upper limit threshold is determined by collecting multiple sets of reference extreme value amplitude ratios under historical extreme safety test environments and extracting the maximum value among the multiple sets of reference extreme value amplitude ratios.

[0012] As a further aspect of the present invention, the feature fusion module includes: The waveform alignment submodule calls the torso vertical acceleration waveform and the preset abnormal reference waveform for the corresponding time period of the force anomaly sequence, performs nonlinear dynamic warping operation in the time dimension, finds the time-series mapping path with the smallest waveform difference between the two sets of sequences, extracts the fluctuation anomaly node at the corresponding position along the time-series mapping path, and generates acceleration mutation point. The time interval calculation submodule calls the acceleration mutation point, extracts the abnormal trigger time contained in the force anomaly sequence, subtracts the abnormal trigger time from the corresponding time mapped by the acceleration mutation point, performs a time dimension subtraction difference calculation operation, calculates the corresponding time span, and obtains the time interval value. The feature vector construction submodule extracts the time interval values ​​for a consecutive preset number of sampling periods for the time interval values, and performs a one-dimensional matrix arrangement and splicing operation on the time interval values ​​according to the time node ordering benchmark rule to generate a critical instability feature vector.

[0013] As a further aspect of the present invention, the state assessment module includes: The risk mapping submodule calls the critical instability feature vector, performs inner product scalar calculation between the critical instability feature vector and the weight normal vector of the pre-set support vector machine model, adds a bias constant, extracts the classification space margin, performs log-odds distribution mapping on the classification space margin, and generates a risk probability value. The status determination submodule compares the risk probability value with a preset alarm threshold for the risk probability value, retains the risk probability value that exceeds the alarm threshold limit, and removes the risk probability value that does not exceed the alarm threshold limit to generate an abnormal risk status quantity. The instruction encapsulation submodule obtains the edge node attribute sequence based on the abnormal risk state quantity, performs field sequence concatenation operation on the edge node attribute sequence and the abnormal risk state quantity according to the IoT communication message architecture, adds a communication redundancy check code segment, and generates a security warning instruction.

[0014] As a further aspect of the present invention, the execution of log-odds distribution mapping refers to using the natural constant as the base, calculating the negative power of the classification space interval, adding one to the calculation result and then taking the reciprocal. The preset alarm threshold is determined by the lower boundary value of the confidence interval obtained by performing statistical distribution fitting calculation on the risk probability values ​​extracted from historical abnormal sample sequences.

[0015] On the other hand, the IoT-based safety monitoring method for high-altitude power operations, which is executed based on the aforementioned IoT-based safety monitoring system for high-altitude power operations, includes the following steps: S1: Obtain seat belt tension data and torso vertical acceleration through IoT gateway, perform time sequence alignment matching based on time records, and construct multi-source sensing time sequence matrix; S2: Based on the multi-source sensing time-series matrix, extract the transition time when the tension difference changes from positive to negative, calculate the front window rate and the back window rate, filter sample points with the front window rate greater than a preset positive threshold and the back window rate less than a preset negative threshold, and generate a rate mutation sequence. S3: Extract the synchronization node difference based on the rate mutation sequence, count the proportion of sign-consistent difference, calculate the extreme value amplitude ratio for sampling points below the preset ratio threshold, extract the time when the extreme value amplitude ratio exceeds the preset upper limit threshold, and construct the force anomaly sequence. S4: The dynamic time warping algorithm is used to align the waveform of the abnormal force sequence with the vertical acceleration of the torso to obtain the acceleration mutation point, calculate the time interval between the acceleration mutation point and the abnormal triggering time, and generate the critical instability feature vector. S5: Call the critical instability feature vector, perform classification operation through the support vector machine model to extract abnormal risk states, and generate safety warning instructions.

[0016] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following: By acquiring tension values ​​and vertical acceleration and performing matching based on time records to generate a multi-source sensing matrix, the bias interference caused by single data acquisition is eliminated. By combining difference extraction and window rate comparison to screen mutation sequences to identify the mutation points of force evolution, the abnormal operation triggering end is locked, thereby breaking through the bottleneck of manual monitoring delay. The difference of synchronization nodes is statistically analyzed and the extreme value amplitude ratio is calculated to construct an abnormal sequence. Microscopic danger signs in normal operation signals are deeply mined. The time warping algorithm is used to align abnormal waveforms to obtain the acceleration mutation interval and establish feature vectors. The temporal delay relationship in the unstable evolution process is accurately quantified. Based on the classification model, early warning instructions are issued under abnormal risk conditions, thus constructing an active risk interception defense line. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the accompanying drawings without creative effort.

[0018] Figure 1 This is a system schematic diagram of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention; Figure 3 This is a flowchart of the state sensing module in this invention; Figure 4 This is a flowchart of the timing analysis module in this invention; Figure 5 This is a flowchart of the gradient determination module in this invention; Figure 6 This is a flowchart of the feature fusion module in this invention; Figure 7 This is a flowchart of the state assessment module in this invention; Figure 8 This is a flowchart of the method of the present invention. Detailed Implementation

[0019] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0020] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0021] This invention provides an IoT-based safety monitoring system for high-altitude power operations, such as... Figure 1-2 The diagram shown illustrates an IoT-based safety monitoring system for high-altitude power operations. The system includes: The state perception module acquires the seat belt tension and torso vertical acceleration values ​​through the IoT gateway, performs time-series alignment matching on the tension and vertical acceleration values ​​based on time records, and constructs a multi-source perception time-series matrix. The time series analysis module extracts the transition time of the tension difference from positive to negative based on the multi-source sensing time series matrix, calculates the front window rate and the back window rate, filters sample points with the front window rate greater than a preset positive threshold and the back window rate less than a preset negative threshold, and generates a rate mutation sequence. The gradient determination module extracts the difference between synchronization nodes based on the rate mutation sequence, counts the proportion of difference with consistent signs, calculates the extreme value amplitude ratio for sampling points below a preset ratio threshold, extracts the moment when the extreme value amplitude ratio exceeds a preset upper limit threshold, and constructs a force anomaly sequence. The feature fusion module uses a dynamic time warping algorithm to perform waveform alignment between the abnormal force sequence and the vertical acceleration of the torso to obtain acceleration abrupt points, calculates the time interval between the acceleration abrupt points and the abnormal triggering time, and generates a critical instability feature vector. The status assessment module calls the critical instability feature vector, performs classification operations through the support vector machine model to map the risk probability value, filters abnormal risk states whose risk probability values ​​exceed the preset alarm threshold, and uses the Internet of Things communication protocol to encapsulate the abnormal risk state and edge node attributes into data packets to generate a security warning instruction.

[0022] The multi-source sensing time series matrix includes timestamps, spatial coordinates, and signal-to-noise ratio; the rate mutation sequence includes peak acceleration, rate of change, and relaxation time; the force anomaly sequence includes the action period, stress amplitude, and off-center load coefficient; the critical instability feature vector includes time delay, phase difference, and cross-correlation coefficient; and the safety warning instructions include interruption priority, linkage control signals, and fault codes.

[0023] Specifically, such as Figure 2 , 3 As shown, the state awareness module includes: The data receiving submodule obtains the seat belt tension and torso vertical acceleration from the IoT gateway, configures the acquisition cycle for the IoT gateway, calibrates the time record of the seat belt tension and torso vertical acceleration according to the acquisition cycle, and extracts the timestamp to obtain the node synchronization time. During the acquisition of seat belt tension and torso vertical acceleration values, a resistance strain gauge tension sensor deployed at the D-ring attachment point on the back of the worker's seat belt, and a microelectromechanical vertical accelerometer built into the back area of ​​the worker's work clothes, continuously perform high-frequency physical quantity acquisition operations. Before receiving actual sensor data, a specific acquisition cycle is configured for the IoT gateway. This acquisition cycle is strictly set based on the physical transient changes in the moment of an accidental fall by a worker at height. The weightlessness reaction and the sudden change in rope force during the initial stage of the fall are extremely short. To ensure accurate capture of high-frequency transient change signals and avoid Nyquist sampling aliasing, the acquisition cycle is precisely configured to 0.02 seconds. In the internal microprocessor core, based on the aforementioned 0.02-second acquisition cycle, the time of each received seat belt tension and torso vertical acceleration value is independently calibrated and recorded. The time recording calibration mechanism uses an internal real-time clock calibrated with a high-precision network time protocol as the absolute reference. It employs a unified Coordinated Universal Time (UTC) format to convert absolute time into a millisecond-level digital sequence. This absolute time millisecond sequence is extracted as a timestamp to obtain the node synchronization time. Regarding the specific parameter assignment and acquisition process, within a specific acquisition cycle, the current seatbelt tension is read as 450 Newtons, and the current torso vertical acceleration is read as 9.8 meters per second squared. At the physical instant these two values ​​are read, the internal real-time clock is retrieved through the underlying register, obtaining the current absolute time millisecond sequence as 1678000000000 milliseconds. This 1678000000000 milliseconds is then established as the node synchronization time for this acquisition. The advantage of this operational logic is that it forcibly assigns an absolute time scale to various heterogeneous sensor data, effectively eliminating the potential for clock drift and timing discrepancies caused by ambient temperature affecting the internal crystal oscillators of individual sensors, significantly improving the accuracy of subsequent alignment processing of multi-source data.

[0024] The timing alignment submodule calls the node synchronization time amount, calculates the difference between the time record of the seat belt tension acquisition value and the time record of the torso vertical acceleration, and determines that when the difference is less than the alignment deviation benchmark value, it filters the numerical items in the corresponding time node and integrates them to generate a status alignment data group. During the timing feature matching phase, the acquired node synchronization time values ​​are retrieved, and the time records associated with the seat belt tension acquisition values ​​and the torso vertical acceleration are extracted. Then, a subtraction operation is performed, subtracting the torso vertical acceleration time record from the seat belt tension acquisition value-associated time record, and taking the absolute value of the subtraction to calculate the difference between these two independent time records. An alignment deviation benchmark value is pre-set in the judgment logic. By extracting data transmission delay logs from real high-altitude operation scenarios over the past 30 days, the delay value at the 95th percentile of the log, 0.005 seconds, is extracted and rigorously set as the alignment deviation benchmark value. The calculated time record difference is compared with this 0.005-second alignment deviation benchmark value. When the time record difference is determined to be strictly less than the 0.005-second alignment deviation benchmark value, it is determined that these two heterogeneous data belong to the same transient action space at the time of physical occurrence. The tension and acceleration values ​​corresponding to these two time nodes are then filtered from the internal buffer memory. The seatbelt tension data and torso vertical acceleration data that meet the above conditions are vertically merged and stitched together in memory. For the specific calculation process, assuming the time record associated with the called seatbelt tension data is 1678000000010 milliseconds and the time record associated with the torso vertical acceleration is 1678000000013 milliseconds, subtracting 1678000000010 milliseconds from 1678000000013 milliseconds yields an absolute difference of 3 milliseconds. Converting 3 milliseconds to 0.003 seconds, this is compared with the alignment deviation benchmark value of 0.005 seconds, determining that 0.003 seconds is less than 0.005 seconds. Based on this result, the numerical items within the corresponding time node are selected, and the tension and acceleration at that moment are packaged to generate a state-aligned data group. The advantage of this operational logic is that it physically eliminates occasional asynchronous data frames caused by wireless communication channel fading or congestion, ensuring strict time synchronization for subsequent time-series analysis feature construction.

[0025] The matrix construction submodule, based on the state-aligned data group, sets the seat belt tension acquisition values ​​as a horizontal element sequence and the torso vertical acceleration as a vertical element sequence. The horizontal and vertical element sequences are stored in a two-dimensional grid structure according to the same time node to establish a multi-source sensing time sequence matrix. Based on the generated continuous state-aligned data set, a contiguous two-dimensional storage space is allocated in the underlying memory controller for structured reassembly. Seatbelt tension data from 200 consecutive acquisition cycles within the state-aligned data set are extracted and linearly arranged according to their absolute temporal order, forming a horizontal element sequence with a fixed length of 200 numerical nodes. Similarly, torso vertical acceleration within the same time span is extracted and also arranged chronologically as a vertical element sequence, also with a length of 200 numerical nodes. The horizontal and vertical element sequences are then stored sequentially into the allocated two-dimensional grid structure according to the same time node index relationship. In this grid structure, the first row continuously stores the tension data sequence, the second row continuously stores the acceleration data sequence, and each column of the matrix corresponds to an absolutely identical synchronization timestamp, thus establishing a standardized multi-source sensing temporal matrix. Regarding the parameter acquisition and matrix combination process, for example, at the first time node, the seat belt tension is 450 Newtons and the torso vertical acceleration is 9.8 m / s². At the second time node, the seat belt tension changes to 455 Newtons, and the torso vertical acceleration is finely adjusted to 9.7 m / s². Through addressing operations, 450 and 455 are filled into the first row and first column of the two-dimensional grid structure, and 9.8 and 9.7 are filled into the second row and first column of the two-dimensional grid structure, respectively. By iterating through 200 time nodes, all data is filled, and finally a multi-source sensing time series matrix with a size of 2 rows and 200 columns is established. Table 1: Multi-source sensing time-series matrix data table Monitoring time sequence number Tension acquisition values Accelerometer data acquisition 1 450 9.8 2 455 9.7 3 460 9.6 4 458 9.8 5 452 9.9 As shown in Table 1, the sensor data from different dimensions were precisely aligned under the same monitoring time sequence number, forming a structured time series matrix, which provides direct and continuous data support for capturing subsequent transition moments.

[0026] Specifically, such as Figure 2 , 4 As shown, the time series analysis module includes: The transition moment extraction submodule extracts tension sampling values ​​based on the multi-source sensing time series matrix, matches time series nodes, calculates the difference between tension sampling values ​​of adjacent nodes, determines the location where the sign of the difference changes from positive to negative, extracts the location-related time record, and generates the transition moment when the tension difference changes from positive to negative. The system scans the pre-established multi-source sensing time-series matrix and sequentially extracts seatbelt tension sampling values ​​from each consecutive time node along the first row of the matrix's storage area. Following a unidirectional temporal sequence, the system extracts the tension sampling value of the second node and subtracts it from the tension sampling value of the first node. This subtraction is used to calculate the difference between the tension sampling values ​​of adjacent nodes, representing the tension change gradient at a microscale. The system monitors and judges the mathematical sign of the continuously calculated difference values ​​in real time. When the currently calculated difference value is negative and the immediately preceding difference value is positive, the system identifies the location where the difference sign changes from positive to negative. The system then locates the next time-series node corresponding to this sign change position and extracts the absolute time record associated with that node in the time-series matrix. This absolute time record is used to generate the moment when the tension difference changes from positive to negative. Regarding the derivation of specific values, assuming the tension value extracted to the 5th node is 460 Newtons, the tension value to the 6th node is 475 Newtons, and the tension value to the 7th node is 465 Newtons. Subtracting the 460 Newtons from the 475 Newtons at the 6th node yields a difference of +15 Newtons (positive sign). Then, subtracting the 475 Newtons from the 465 Newtons at the 7th node yields a difference of -10 Newtons (negative sign). Comparing these two differences, a physical reversal event is determined, where the positive value turns negative. This event occurs precisely at the transition point from the 6th to the 7th node. The absolute time record of the 7th node, 1678000000120 milliseconds, is extracted, generating the transition time of the tension difference from positive to negative at 1678000000120 milliseconds. The advantage of this operational logic is that it can accurately capture the physical inflection point of the seat belt's stress transitioning from tension and tightness to elastic relaxation with extremely low microcontroller computing power consumption, without requiring complex frequency domain transformations, thus providing a precise cutting anchor point for subsequent rate calculations.

[0027] The window rate calculation submodule calls the moment when the tension difference changes from positive to negative. It uses the moment when the tension difference changes from positive to negative to extract a fixed time period on both sides of the time axis to define the pre-calculation window and the post-calculation window. It calculates the pre-window rate and the post-window rate by dividing the tension change within the window by the time span and obtains the window rate group. The moment when the tension difference generated during the call process changes from positive to negative is used as the absolute timestamp of this transition point as the central reference point. Fixed time intervals are symmetrically extracted on both sides of the time axis in the past and future directions. The fixed time interval parameter is precisely configured as 0.5 seconds. The 0.5-second interval in the past direction is defined as the pre-calculation window, and the 0.5-second interval in the future direction is defined as the post-calculation window. The tension value at the end of the pre-calculation window is extracted, and the tension value at the beginning of the pre-calculation window is subtracted to obtain the absolute change in tension within the pre-calculation window. This absolute change in tension within the pre-calculation window is divided by the fixed time span of 0.5 seconds to calculate the pre-calculation window rate. Similarly, the tension value at the end of the post-calculation window is extracted, and the tension value at the beginning of the post-calculation window is subtracted to obtain the tension change within the post-calculation window. This value is divided by the time span of 0.5 seconds to calculate the post-calculation window rate. The two are combined to obtain the window rate group. Regarding the specific calculation assignment and physical mapping process, assuming the transition time of the called tension difference from positive to negative is 1678000000120 milliseconds, and the starting time of the preceding calculation window is 1678000000020 milliseconds, the tension read at this starting time is 400 Newtons, while at the end of the transition point, the tension surges to 2400 Newtons. Subtracting 400 Newtons from 2400 Newtons yields a tension change of 2000 Newtons. Dividing 2000 Newtons by 0.5 seconds gives the preceding window rate of 4000 Newtons per second. Next, the tension at the beginning of the following calculation window is 2400 Newtons, and at the end, the tension drops back to 800 Newtons. Subtracting 2400 Newtons from 800 Newtons yields a change of -1600 Newtons. Dividing this by 0.5 seconds gives the following window rate of -3200 Newtons per second. This generates a window rate group containing 4000 Newtons per second and -3200 Newtons per second. The advantage of this operational logic is that it objectively quantifies the transient evolution intensity of the seat belt's mechanical state before and after the impact, eliminating the interference of local minor noise caused by high-frequency sampling on the overall force trend analysis.

[0028] The mutation sequence screening submodule, based on the rate evolution trend of the window rate group comparison, determines that the preceding window rate is greater than a preset positive threshold and the following window rate is less than a preset negative threshold, screens test sample points associated with window rate groups that conform to the evolution trend, merges test sample points according to time sequence, and establishes rate mutation sequences. For the generated window rate groups, a rigorous comparative analysis of the rate evolution trends is performed on the values ​​within them. The preset positive threshold is determined by extracting the lower boundary value of the tension increase rate statistics during the suspension impact phase in historical extreme drop simulation tests; the preset negative threshold is determined by extracting the upper boundary value of the tension decay rate statistics during the impact rebound phase in the same historical extreme drop simulation tests. The preceding window rate in the window rate group is extracted, and its value is checked against the preset positive threshold. Subsequently, the following window rate is extracted, and its value is checked against the preset negative threshold. When both the preceding window rate and the following window rate are confirmed to be greater than the preset positive threshold and less than the preset negative threshold, the window rate group is determined to fully conform to the typical physical trend of high-altitude drop suspension impact and rebound evolution. All original tension test sample points associated with this window rate group conforming to the evolution trend are selected. The selected test sample points are stored in a high-speed cache, and are sequentially merged and sorted according to the timestamps of each sample point to establish a continuous and pure rate mutation sequence. In the specific logical calculation, assuming the system sets a preset positive threshold of 2000 N / s and a preset negative threshold of -1500 N / s based on historical simulation test data, when retrieving a window rate group containing a pre-window rate of 4000 N / s and a post-window rate of -3200 N / s, it first determines that 4000 N / s is greater than the preset positive threshold (i.e., 4000 is greater than 2000), and then determines that -3200 N / s is less than the preset negative threshold (i.e., -3200 is less than -1500). Based on this determination, all 50 test sample points obtained within a total time span of 1 second covered by the pre- and post-calculation windows, based on a sampling period of 0.02 seconds, are retrieved and seamlessly stitched together in memory in ascending order of timestamps, ultimately establishing a rate mutation sequence containing 50 numerical nodes. The advantage of this operational logic is that it accurately filters out unidirectional, slowly varying tension fluctuations or pseudo-abrupt changes caused by workers climbing, turning, or lifting heavy objects, ensuring that the data input into the subsequent deep feature extraction stage has extremely high physical purity.

[0029] Specifically, such as Figure 2 , 5 As shown, the gradient determination module includes: The difference percentage statistics submodule extracts the synchronization node difference based on the rate mutation sequence, performs sign consistency check on the synchronization node difference, calculates the proportion of the number of sign-consistent differences to the total number of all records, and generates the sign-consistent difference percentage. Based on the continuous rate mutation sequence established in the previous stage, the numerical difference between tension nodes at every two adjacent synchronization moments in the sequence is extracted. A strict sign consistency check is performed on each extracted synchronization node difference. Specifically, the positive and negative mathematical signs of all calculated differences are extracted, the distribution of positive and negative signs is statistically analyzed, and the sign direction that accounts for the majority is selected as the reference sign for the sequence. Then, all difference records are traversed, and the absolute number of differences whose signs are exactly the same as the reference sign is counted. The number of differences with consistent signs is used as the numerator, divided by the total number of adjacent differences recorded in the rate mutation sequence as the denominator, to calculate the percentage of differences with consistent signs out of the total number of records, generating the percentage of differences with consistent signs. Taking the extrapolation of actual monitoring data as an example, assuming that continuous tension values ​​at 50 time points are extracted from a rate mutation sequence during a fall decay period, the tension change difference between 49 adjacent nodes is calculated by subtracting them pairwise. Of these, 45 difference calculation results are negative, representing a continuous decline in tension, and only 4 difference calculation results are positive. Negative numbers were used as the baseline sign, and 45 discrepancies with matching signs were identified, out of a total of 49. Dividing 45 by 49 yielded a matching sign difference ratio of 0.918. This calculated 0.918 was then output as the result to the next threshold comparison stage. The advantage of this calculation logic is that it quantifies the unidirectional concentration of tension attenuation during the stress release stage in a dimensionless manner, providing a relative physical attenuation state assessment index that is unaffected by the absolute magnitude of a person's weight.

[0030] The amplitude ratio calculation submodule compares the proportion of sign consistency difference with the preset ratio threshold. For time-series sampling points whose values ​​are lower than the preset ratio threshold, it extracts the local tension extreme value and the normal operating amplitude. The local tension extreme value is divided by the normal operating amplitude to obtain the extreme value amplitude ratio. The system receives the percentage of sign-consistency differences calculated in the previous stage and compares it with a preset threshold in memory. This preset threshold is set to 0.65. It then determines whether the currently input percentage of sign-consistency differences is lower than this threshold. If the result is lower than 0.65, it retrieves the original data collected within the corresponding time period, extracts the highest peak value of tension during this period as the local tension extreme value, and retrieves the average stress tension of the person under normal operating static suspension conditions as the normal operating amplitude. A division operation is performed, dividing the local tension extreme value by the normal operating amplitude to calculate the extreme value amplitude percentage. In the actual calculation and comparison process, the percentage of sign-consistency differences generated in the previous stage is 0.55. This is compared with the preset threshold of 0.65, and since 0.55 is less than 0.65, the extreme value calculation condition is triggered. The recorded local tension extreme value is extracted as 3000 Newtons, and the normal operating amplitude is extracted as 600 Newtons. Dividing 3000 by 600 yields a result of 5, which is then determined as the extreme value amplitude ratio. The advantage of this calculation logic is that it effectively eliminates the differences in static baseline caused by the weight of workers and the weight of their equipment, achieving a highly efficient normalization conversion of absolute force level into relative impact intensity.

[0031] Table 2: Statistics of Extreme Value Amplitude Ratio Sample batch number Percentage of differences with consistent signs Local tension extremes Normal operating amplitude Extreme value amplitude ratio 1 0.55 3000 600 5.0 2 0.58 2400 600 4.0 3 0.42 4200 700 6.0 4 0.61 2200 550 4.0 5 0.49 3500 500 7.0 As shown in Table 2, by associating the proportion of the sign consistency difference of the triggering abnormal conditions and substituting the corresponding local extreme values ​​and individual normal amplitudes, the proportion coefficient reflecting the relative impact intensity was calculated. The batch coefficient value intuitively reflects the true degree of force abnormality after removing the interference of body weight.

[0032] The abnormal sequence construction submodule extracts the target time nodes corresponding to the extreme value amplitude ratios that exceed the preset upper limit threshold, and performs merging and arrangement of the target time nodes according to the time sequence to construct the stress anomaly sequence. For the output extreme value amplitude ratio, its corresponding specific value is extracted and compared with an internally loaded preset upper limit threshold. Extreme value amplitude ratios exceeding the preset upper limit threshold are selected. This preset upper limit threshold is set to 4.5, reflecting the boundary of catastrophic impact. When the currently input extreme value amplitude ratio is determined to be greater than 4.5, all time-series data frames around the target time node corresponding to this extreme value amplitude ratio are extracted. The extracted target time nodes are merged and arranged according to their absolute occurrence time order, connecting the discrete high-risk fluctuation nodes end-to-end in the cache to construct a complete stress anomaly sequence. In a specific example demonstration, the received extreme value amplitude ratio calculated in the previous sequence is 5. A rigorous comparison of 5 with the preset upper limit threshold of 4.5 is performed, determining that 5 is greater than 4.5. Based on the judgment result of exceeding the limit threshold, all high-frequency data of the 30 target time nodes within the entire impact burst time window that generated the coefficient 5 are extracted and linked sequentially from front to back according to the timestamp. Irrelevant stationary time gaps are removed, and finally a highly focused stress anomaly sequence covering these 30 nodes is constructed in memory. The advantage of this operation logic is that it ensures that the data frames finally included in the anomaly sequence reflect real high-risk impacts that are sufficient to cause structural failure or damage to internal organs, and significantly reduces the data scale that needs to be subjected to subsequent complex regularization operations.

[0033] Specifically, such as Figure 2 , 6 As shown, the feature fusion module includes: The waveform alignment submodule calls the torso vertical acceleration waveform and the preset abnormal reference waveform for the corresponding time period of the force anomaly sequence, performs nonlinear dynamic warping operation in the time dimension, finds the time-series mapping path with the smallest waveform difference between the two sets of sequences, extracts the corresponding fluctuation anomaly node along the time-series mapping path, and generates acceleration mutation point. The system invokes the previously constructed high-risk stress anomaly sequence and retrieves the torso vertical acceleration waveform and the preset anomaly reference waveform, which correspond perfectly to the time span of the stress anomaly sequence, from the memory buffer. For these two sets of local oscillation frequency sequences with the same physical dimensions but exhibiting individual motion differences and mechanical transmission delays, a nonlinear dynamic warping operation in the time dimension is performed. First, the Euclidean absolute difference between each actual acceleration node in the torso vertical acceleration waveform and each reference acceleration node in the preset anomaly reference waveform is calculated, constructing a large-scale two-dimensional distance matrix. Then, starting from the upper left corner of this two-dimensional distance matrix, based on the dynamic programming principle of minimizing the cumulative distance between adjacent grid nodes, the system searches for the path with the minimum cumulative distance, grid by grid towards the lower right corner, to find the temporal mapping path with the minimum waveform difference between the two sets of sequences. Along this determined temporal mapping path, each matching coordinate pair on the path is traversed, and the torso vertical acceleration node corresponding to the position representing a significant jump in weightlessness in the preset anomaly reference waveform is extracted. This node is defined as a fluctuation anomaly node indicating physical imbalance. For the pathfinding process of specific parameters, a 30-node vertical acceleration waveform of the torso and a 35-node preset abnormal reference waveform are acquired. After calculating a 30x35 two-dimensional distance matrix, a nonlinear temporal mapping path containing 40 matching coordinate pairs is derived using a dynamic programming algorithm. The 15th matching coordinate pair on the path where the reference waveform exhibits a sudden change in weightlessness is extracted. At this point, the reference acceleration value of the preset abnormal reference waveform at this coordinate drops abruptly from 9.8 m / s² to around 0.5 m / s². The time node data corresponding to the vertical acceleration of the torso in this coordinate pair is read, and it is found that the actual vertical acceleration of the torso at the corresponding moment also changes from 9.8 m / s² to an extreme value of 0.5 m / s². The actual moment node where the abrupt weightlessness change of 0.5 m / s² occurs is extracted to generate the acceleration mutation point. The advantage of this operational logic is that it effectively eliminates the mechanical elastic physical hysteresis effect generated when the stress deformation is transmitted to the human torso workwear, and realizes the matching and positioning of stress pulse occurrence and human posture weightlessness displacement in the absolute time dimension.

[0034] The time interval calculation submodule calls the acceleration mutation point, extracts the abnormal trigger time contained in the force anomaly sequence, subtracts the abnormal trigger time from the corresponding time mapped by the acceleration mutation point, performs a time dimension subtraction difference calculation operation, calculates the corresponding time span, and obtains the time interval value. The process involves calling the acceleration mutation point generated by the mapping and simultaneously accessing the data header structure of the force anomaly sequence to extract the anomaly trigger time contained in the force anomaly sequence. This anomaly trigger time is the absolute timestamp of the start when the tension begins to exceed the safety warning baseline and experience a sharp increase. Using the corresponding actual occurrence time mapped from the acceleration mutation point as the minuend and the anomaly trigger time as the subtrahend, a time-dimensional subtraction difference calculation operation is performed, subtracting the anomaly trigger time from the corresponding time. Through this absolute subtraction operation in the time dimension, the time span corresponding to the delay experienced from the onset of force anomaly in the fall arrest equipment to the transfer of physical kinetic energy to the human torso, resulting in a fundamental weightlessness mutation in acceleration, is accurately calculated, yielding the time interval value. In the calculation example, the anomaly trigger time of the force anomaly sequence extracted from the log is 1678000000150 milliseconds, while the corresponding time of the acceleration mutation point mapping obtained after normalization is 1678000000270 milliseconds. A time-difference subtraction operation is performed, subtracting 1678000000270 milliseconds from 1678000000150 milliseconds, resulting in 120 milliseconds. This 120 milliseconds is explicitly calculated as the delay time span corresponding to energy conduction, ultimately yielding a time interval value of 120 milliseconds. This value is temporarily stored in a dedicated feature register for subsequent high-dimensional feature aggregation and assembly.

[0035] The feature vector construction submodule extracts the time interval values ​​for a continuous preset number of sampling periods, and performs a one-dimensional matrix arrangement and splicing operation on the time interval values ​​according to the time node ordering benchmark rule to generate critical instability feature vectors. Based on the time interval values ​​calculated and output in the preceding stage, a multi-dimensional feature aggregation and assembly process is initiated. A preset number of sampling periods is set in the memory queue, strictly defined as 10 sampling periods. Ten dynamically generated time interval values ​​are continuously extracted within these 10 sampling periods. Following a time node chronological order (strictly adhering to the physical order of the underlying acquisition timestamps from smallest to largest), a one-dimensional matrix arrangement and splicing operation is performed on these ten independent time interval values, integrating the scalar values ​​originally scattered across different time slices into a single structured feature array. For the specific integration, arrangement, and splicing operation, within the time window including the initial impact and rebound, the following ten dynamically changing time interval values ​​are collected sequentially: 120 milliseconds, 125 milliseconds, 118 milliseconds, 130 milliseconds, 135 milliseconds, 132 milliseconds, 138 milliseconds, 140 milliseconds, 145 milliseconds, and 142 milliseconds. Following the chronological order of the time points, these 10 values ​​are used as fixed-dimensional elements to fill specific slots in a one-dimensional array structure, forming an ordered sequence of 10 elements. This sequence is then encapsulated and output; this sequence is the generated critical instability feature vector representing the complete fall damping characteristics. The advantage of this operational logic is that it preserves to the greatest extent possible the low-frequency oscillations and damping attenuation characteristics of the fall impact repeatedly transmitted between the human body and elastic protective equipment, providing a highly information-dense input feature carrier for downstream complex nonlinear classification and mapping operations.

[0036] Specifically, such as Figure 2 , 7 As shown, the status assessment module includes: The risk mapping submodule calls the critical instability feature vector, performs inner product scalar calculation between the critical instability feature vector and the weight normal vector of the pre-set support vector machine model, adds a bias constant, extracts the classification space margin, performs log-odds distribution mapping on the classification space margin, and generates a risk probability value. The critical instability feature vector generated by the previous integration is invoked and input into the pre-configured classification operation for spatial distance calculation. The 10-dimensional weight normal vector, obtained through training and optimization, is extracted. Each element of the critical instability feature vector is multiplied one by one by the corresponding value in the pre-configured weight normal vector, and all product results are accumulated. An inner product scalar calculation operation is then performed to obtain the inner product result. Subsequently, the bias constant in the calculation logic is extracted, and the inner product result is added to this bias constant. The sum is then used as the classification spatial margin. Next, a log-odds distribution mapping operation is performed on this classification spatial margin. Specifically, the natural constant 2.718 is used as the base, and the negative value of the classification spatial margin is used as the exponent. The negative power of the natural constant base is calculated. The result of this negative power calculation is added with the number 1, and then the arithmetic reciprocal of the sum is taken. This continuous non-linear transformation generates the risk probability value. In a specific numerical calculation demonstration, assuming that the received critical instability feature vector and weight normal vector undergo inner product scalar calculation, the resulting cumulative inner product is 3.5, and the extracted bias constant for translation is 1.0, adding 1.0 to 3.5 yields 4.5, which is the extracted classification space margin. Then, using the natural constant 2.718 as the base, the result is calculated to the power of -4.5, approximately 0.011. Adding 1 to 0.011 yields 1.011, and finally, the reciprocal of 1.011 is taken, resulting in a risk probability of approximately 0.989.

[0037] Table 3: Risk Probability Distribution and State Mapping Table Evaluation serial number Classification Spacing Negative power calculation results Add one and count the reciprocal result Risk probability value 1 4.5 0.011 1.011 0.989 2 2.5 0.082 1.082 0.924 3 -2.0 7.389 8.389 0.119 As shown in Table 3, by strictly following the log-probability mapping operation process, the classification space interval of the distance hyperplane in the high-dimensional space is accurately transformed into a risk probability value within the standard range. The high interval corresponds to an extremely high fall probability, providing a standardized scalar input for the direct truncation judgment of the subsequent alarm threshold.

[0038] The status determination submodule compares the risk probability value with the preset alarm threshold, retains the risk probability values ​​that exceed the alarm threshold, and removes the risk probability values ​​that do not exceed the alarm threshold, thus generating an abnormal risk status quantity. The risk probability value generated by the previous nonlinear mapping is introduced into a comparison process with a preset alarm threshold. The lower boundary value of the confidence interval from past statistical fitting is extracted as 0.88, thus 0.88 is determined as the preset alarm threshold for determining extreme danger. The currently input risk probability value is directly compared with this 0.88. Based on the comparison result, if the current risk probability value is found to be greater than or equal to 0.88 during the judgment process, this value is retained in the high-speed emergency alarm cache. Conversely, if the current risk probability value is found to be less than 0.88 during the judgment process, this risk probability value that does not meet the alarm threshold is directly removed from the memory queue. For the retained data exceeding the threshold, it is encapsulated and transformed to generate an abnormal risk state quantity representing that the current operator is in an extremely dangerous situation requiring immediate intervention. The actual calculated value is substituted for verification. Upon receiving the previously generated risk probability value of 0.989, a comparison operation is performed between 0.989 and the preset alarm threshold of 0.88. The judgment result shows that 0.989 is greater than 0.88, so 0.989 is immediately retained, and an abnormal risk state quantity with an emergency status code of 1 is generated accordingly.

[0039] The instruction encapsulation submodule obtains the edge node attribute sequence based on the abnormal risk state quantity, performs field sequence concatenation operation on the edge node attribute sequence and the abnormal risk state quantity according to the IoT communication message architecture, adds a communication redundancy check code segment, and generates a security warning instruction. Based on the abnormal risk status output from the previous stage, the underlying assembly and radio frequency (RF) transmission mechanism for alarm commands is initiated. First, the internal registration configuration flash table is accessed to obtain the attribute sequence of the edge nodes deployed on-site. This edge node attribute sequence contains a fixed set of text and numerical features, including the hardware media access control address identifier, the physical platform number for the high-altitude operation, and the personnel's real-name identification badge code. The extracted edge node attribute sequence and the received abnormal risk status are then strictly processed according to the standard communication message architecture issued by the low-power wide-area IoT platform. Binary concatenation of the field sequence is performed in the effective data payload area of ​​the message protocol. After concatenating the main information data, a cyclic redundancy check (CRC) algorithm is used to divide the entire concatenated message payload polynomial. The modulo-2 remainder is extracted and used as a communication redundancy check code segment for error detection. This communication redundancy check code segment is seamlessly appended to the end of the concatenated field, ultimately generating a complete command that can be directly modulated and transmitted by the underlying communication baseband. In the specific calculation of parameter combinations, the edge node attribute sequence includes hardware identifier 2005 and operation location number 802. The obtained abnormal risk status is the highest emergency status code 1. 2005, 802, and 1 are seamlessly concatenated into a combined data field 200508021 according to the communication message architecture. A polynomial cyclic redundancy check (CRC) operation is performed on 200508021, yielding an anti-interference check code of 9. This 9 is appended to the end of the combined field, generating the complete numerical sequence 2005080219. This 2005080219 is the final generated safety warning command, which will soon be transmitted to the antenna. The advantage of this operation logic is that it improves the anti-interference capability of emergency warning commands in the harsh radio electromagnetic environment surrounding large metal construction facilities, ensuring that rescue signals can penetrate interference and reach the ground control center.

[0040] Please see Figure 8 The IoT-based safety monitoring method for high-altitude power operations is implemented based on the aforementioned IoT-based safety monitoring system for high-altitude power operations, and includes the following steps: S1: Obtain seat belt tension data and torso vertical acceleration through IoT gateway, perform time sequence alignment matching based on time records, and construct multi-source sensing time sequence matrix; S2: Based on the multi-source sensing time-series matrix, extract the transition time of the tension difference from positive to negative, calculate the front window rate and the back window rate, filter sample points with the front window rate greater than the preset positive threshold and the back window rate less than the preset negative threshold, and generate a rate mutation sequence. S3: Extract the difference of synchronization nodes based on the rate mutation sequence, count the proportion of sign consistency difference, calculate the extreme value amplitude ratio for sampling points below the preset ratio threshold, extract the time when the extreme value amplitude ratio exceeds the preset upper limit threshold, and construct the force anomaly sequence; S4: The dynamic time warping algorithm is used to align the waveforms of the abnormal force sequence and the vertical acceleration of the torso to obtain the acceleration mutation point, calculate the time interval between the acceleration mutation point and the abnormal triggering time, and generate the critical instability feature vector. S5: Call the critical instability feature vector, perform classification operation through the support vector machine model to extract abnormal risk states, and generate safety warning instructions.

[0041] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A power high-altitude operation safety monitoring system based on the Internet of Things, characterized in that, The system includes: The state perception module acquires the seat belt tension and torso vertical acceleration through the IoT gateway, performs time sequence alignment matching based on the time record, and constructs a multi-source perception time sequence matrix. The time series analysis module extracts the transition time of the tension difference from positive to negative based on the multi-source sensing time series matrix, calculates the front window rate and the back window rate, filters sample points with the front window rate greater than a preset positive threshold and the back window rate less than a preset negative threshold, and generates a rate mutation sequence. The gradient determination module extracts the synchronization node difference based on the rate mutation sequence, calculates the proportion of sign-consistent difference, calculates the extreme value amplitude ratio for sampling points below a preset ratio threshold, extracts the moment when the extreme value amplitude ratio exceeds a preset upper limit threshold, and constructs a force anomaly sequence. The feature fusion module uses a dynamic time warping algorithm to align the force anomaly sequence with the vertical acceleration of the torso to obtain acceleration abrupt change points, calculates the time interval between the acceleration abrupt change points and the anomaly trigger time, and generates a critical instability feature vector. The state assessment module calls the critical instability feature vector, performs classification operations through a support vector machine model to extract abnormal risk states, and generates safety warning instructions.

2. The IoT-based power high-altitude operation safety monitoring system according to claim 1, characterized in that, The multi-source sensing time series matrix includes timestamps, spatial coordinates, and signal-to-noise ratio; the rate mutation sequence includes peak acceleration, rate of change, and relaxation time; the force anomaly sequence includes the action period, stress amplitude, and off-center load coefficient; the critical instability feature vector includes time delay, phase difference, and cross-correlation coefficient; and the safety warning command includes interruption priority, linkage control signal, and fault code.

3. The IoT-based power high-altitude operation safety monitoring system according to claim 1, characterized in that, The state awareness module includes: The data receiving submodule obtains the seat belt tension and torso vertical acceleration from the IoT gateway, configures the acquisition cycle for the IoT gateway, calibrates the time record of the seat belt tension and torso vertical acceleration according to the acquisition cycle, and extracts the timestamp to obtain the node synchronization time. The timing alignment submodule calls the node synchronization time amount, calculates the difference between the seat belt tension acquisition value associated time record and the torso vertical acceleration associated time record, and determines that when the difference is less than the alignment deviation benchmark value, it filters the value items in the corresponding time node and integrates them to generate a state alignment data group. The matrix construction submodule, based on the state-aligned data group, sets the seat belt tension acquisition values ​​as a horizontal element sequence and the torso vertical acceleration as a vertical element sequence. The horizontal and vertical element sequences are stored in a two-dimensional grid structure according to the same time node to establish a multi-source sensing time sequence matrix.

4. The IoT-based safety monitoring system for high-altitude power operations according to claim 1, characterized in that, The time series analysis module includes: The transition time extraction submodule extracts tension sampling values ​​based on the multi-source sensing time series matrix, matches time series nodes, calculates the difference between tension sampling values ​​of adjacent nodes, determines the location where the sign of the difference changes from positive to negative, extracts the location-related time record, and generates the transition time when the tension difference changes from positive to negative. The window rate calculation submodule calls the moment when the tension difference changes from positive to negative, and uses the moment when the tension difference changes from positive to negative to extract a fixed time period on both sides of the time axis to define the pre-calculation window and the post-calculation window. Based on the tension change within the window divided by the time span, the pre-window rate and the post-window rate are calculated to obtain the window rate group. The mutation sequence screening submodule, based on the rate evolution trend of the window rate group, determines that the preceding window rate is greater than a preset positive threshold and the following window rate is less than a preset negative threshold, filters test sample points associated with window rate groups that conform to the evolution trend, merges test sample points according to time sequence, and establishes a rate mutation sequence.

5. The IoT-based power high-altitude operation safety monitoring system according to claim 1, characterized in that, The gradient determination module includes: The difference percentage statistics submodule extracts the synchronization node difference based on the rate mutation sequence, performs sign consistency check on the synchronization node difference, calculates the proportion of the number of sign-consistent differences to the total number of all records, and generates the sign-consistent difference percentage. The amplitude ratio calculation submodule compares the sign consistency difference ratio with a preset ratio threshold. For time-series sampling points whose values ​​are lower than the preset ratio threshold, it extracts the local tension extreme value and the normal operating amplitude. The local tension extreme value is divided by the normal operating amplitude to obtain the extreme value amplitude ratio. The abnormal sequence construction submodule extracts the target time nodes corresponding to the extreme value amplitude ratios that exceed the preset upper limit threshold, and performs merging and arrangement of the target time nodes according to the time sequence to construct the stress anomaly sequence.

6. The IoT-based safety monitoring system for high-altitude power operations according to claim 5, characterized in that, The preset proportion threshold is determined by extracting the proportion of consistent differences of multiple sets of reference symbols within a historical normal operating period and calculating the arithmetic mean of the proportions of consistent differences of multiple sets of reference symbols. The preset upper limit threshold is determined by collecting multiple sets of reference extreme value amplitude ratios under historical extreme safety test environments and extracting the maximum value among the multiple sets of reference extreme value amplitude ratios.

7. The IoT-based safety monitoring system for high-altitude power operations according to claim 1, characterized in that, The feature fusion module includes: The waveform alignment submodule calls the torso vertical acceleration waveform and the preset abnormal reference waveform for the corresponding time period of the force anomaly sequence, performs nonlinear dynamic warping operation in the time dimension, finds the time-series mapping path with the smallest waveform difference between the two sets of sequences, extracts the fluctuation anomaly node at the corresponding position along the time-series mapping path, and generates acceleration mutation point. The time interval calculation submodule calls the acceleration mutation point, extracts the abnormal trigger time contained in the force anomaly sequence, subtracts the abnormal trigger time from the corresponding time mapped by the acceleration mutation point, performs a time dimension subtraction difference calculation operation, calculates the corresponding time span, and obtains the time interval value. The feature vector construction submodule extracts the time interval values ​​for a consecutive preset number of sampling periods for the time interval values, and performs a one-dimensional matrix arrangement and splicing operation on the time interval values ​​according to the time node ordering benchmark rule to generate a critical instability feature vector.

8. The IoT-based safety monitoring system for high-altitude power operations according to claim 1, characterized in that, The status assessment module includes: The risk mapping submodule calls the critical instability feature vector, performs inner product scalar calculation between the critical instability feature vector and the weight normal vector of the pre-set support vector machine model, adds a bias constant, extracts the classification space margin, performs log-odds distribution mapping on the classification space margin, and generates a risk probability value. The status determination submodule compares the risk probability value with a preset alarm threshold for the risk probability value, retains the risk probability value that exceeds the alarm threshold limit, and removes the risk probability value that does not exceed the alarm threshold limit to generate an abnormal risk status quantity. The instruction encapsulation submodule obtains the edge node attribute sequence based on the abnormal risk state quantity, performs field sequence concatenation operation on the edge node attribute sequence and the abnormal risk state quantity according to the IoT communication message architecture, adds a communication redundancy check code segment, and generates a security warning instruction.

9. The IoT-based safety monitoring system for high-altitude power operations according to claim 8, characterized in that, The aforementioned log-odds distribution mapping refers to using the natural constant as the base, calculating the negative power of the classification space interval, adding one to the calculation result, and then taking the reciprocal. The preset alarm threshold is determined by the lower boundary value of the confidence interval obtained by performing statistical distribution fitting calculation on the risk probability values ​​extracted from historical abnormal sample sequences.

10. A method for safety monitoring of high-altitude power operations based on the Internet of Things, characterized in that, The power high-altitude operation safety monitoring system based on the Internet of Things as described in any one of claims 1-9 includes the following steps: S1: Obtain seat belt tension data and torso vertical acceleration through IoT gateway, perform time sequence alignment matching based on time records, and construct multi-source sensing time sequence matrix; S2: Based on the multi-source sensing time-series matrix, extract the transition time when the tension difference changes from positive to negative, calculate the front window rate and the back window rate, filter sample points with the front window rate greater than a preset positive threshold and the back window rate less than a preset negative threshold, and generate a rate mutation sequence. S3: Extract the synchronization node difference based on the rate mutation sequence, count the proportion of sign-consistent difference, calculate the extreme value amplitude ratio for sampling points below the preset ratio threshold, extract the time when the extreme value amplitude ratio exceeds the preset upper limit threshold, and construct the force anomaly sequence. S4: The dynamic time warping algorithm is used to align the waveform of the abnormal force sequence with the vertical acceleration of the torso to obtain the acceleration mutation point, calculate the time interval between the acceleration mutation point and the abnormal triggering time, and generate the critical instability feature vector. S5: Call the critical instability feature vector, perform classification operation through the support vector machine model to extract abnormal risk states, and generate safety warning instructions.

Images