Plateau mining worker health risk early warning method based on multi-modal data

Abstract

The invention relates to the technical field of occupational health monitoring and safety early warning, and discloses a plateau mining employee health risk early warning method based on multi-modal data, and the method comprises the steps: firstly collecting the physiological, movement and environmental data of mining employees, carrying out the time-space alignment processing, and constructing an effective hypoxia work index; secondly, extracting dual-channel physiological response characteristics representing a force-electricity and force-chemical coupling mechanism, and mapping the dual-channel physiological response characteristics to a two-dimensional state space to generate a hysteresis loop; then calculating a centroid drift vector and a coupling coherence coefficient of a hysteresis loop relative to a resting reference state, and constructing a comprehensive health risk index; and finally, executing graded early warning based on a comparison result of the index and a preset safety threshold and a coupling state. According to the method, by quantifying the dynamic response of the physiological system to physical work of the low-oxygen environment and the collaborative state between the systems, high-load fatigue and compensation function failure are effectively distinguished, and the accuracy of plateau operation early warning is improved.

Description

Technical Field

[0001] This invention relates to the field of occupational health monitoring and safety early warning technology, specifically a method for early warning of health risks for workers in plateau mining industries based on multimodal data. Background Technology

[0002] The working environment in high-altitude mines is characterized by low air pressure, low oxygen partial pressure, and extreme cold. Workers face severe physiological challenges when performing physical labor in this environment. To ensure worker safety, wearable smart devices are commonly used to collect data such as heart rate, blood oxygen saturation, and exercise acceleration. Health risks are assessed by monitoring whether these values ​​exceed safe limits.

[0003] However, existing health early warning technologies still have limitations in practical applications. Current monitoring methods often treat the intensity of physical work separately from environmental factors, frequently applying exercise load assessment standards from plains areas or monitoring only environmental parameters independently. This approach ignores the non-linear impact of the low-oxygen environment at high altitudes on the efficiency of human work; that is, under low oxygen partial pressure conditions, generating the same physical acceleration often requires consuming more physiological reserves, making it impossible to objectively reflect the actual physiological burden borne by workers based solely on exercise data.

[0004] Furthermore, current early warning strategies largely rely on static threshold judgments of instantaneous physiological indicators. The human physiological system is a complex system with inertia and regulatory lag; physiological responses are delayed relative to changes in physical load. Simply relying on the judgment of exceeding limits in instantaneous values ​​makes it difficult to capture the dynamic characteristics and cumulative effects of physiological regulation processes, and cannot accurately assess the degree to which the physiological system deviates from steady state.

[0005] More critically, existing technologies typically lack the ability to analyze the synergistic relationships between different physiological subsystems. Under high-intensity work conditions, elevated heart rate and decreased blood oxygen levels can be both normal physiological stress responses and pathological signs of impending failure of the body's compensatory functions. Existing methods struggle to effectively distinguish between positively synergistic high-load fatigue states and anisotropic decoupling compensatory failure states, easily leading to false alarms or missed alarms, and failing to meet the need for precise classification and early warning of health risks in complex high-altitude work environments. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a method for early warning of health risks for miners in high-altitude mining areas based on multimodal data. This method aims to solve the problems in existing technologies, such as the difficulty in objectively quantifying the actual physiological costs under high-altitude hypoxic conditions, the lack of analytical capabilities for dynamic responses and steady-state deviations of physiological systems, and the inability to effectively distinguish between normal high-load fatigue and the failure of physiological compensatory functions, which leads to insufficient accuracy in early warning.

[0007] To achieve the above objectives, this invention provides a method for early warning of health risks for workers in plateau mining operations based on multimodal data. This method includes: collecting physiological and exercise data of mining workers, as well as environmental data from the work site; performing spatiotemporal alignment preprocessing on the collected data; constructing an effective hypoxia work index based on physical work intensity and environmental oxygen partial pressure, using it as a common independent variable for hysteresis loop analysis; extracting dual-channel physiological response features from the spatiotemporally aligned data, constructing cardiovascular response features characterizing the electromechanical coupling mechanism and metabolic response features characterizing the mechanochemical coupling mechanism, respectively; using a sliding time window, mapping the effective hypoxia work index to the cardiovascular and metabolic response features respectively onto a two-dimensional state space, generating a mechanocardial hysteresis loop and a mechanometabolic hysteresis loop, and calculating the centroid drift vector of each hysteresis loop relative to the resting baseline state; calculating the coupling coherence coefficient based on the centroid drift vectors of the two hysteresis loops; constructing a comprehensive health risk index by combining the degree of vector anisotropy and the energy loss of the hysteresis loops; and performing graded early warning decisions based on the comparison results between the comprehensive health risk index and a preset safety threshold.

[0008] Preferably, spatiotemporal alignment preprocessing is achieved by establishing a multi-channel data buffer queue and performing system synchronization. The specific process includes: assigning global time tags to heterogeneous data streams arriving at the buffer queue based on a unified system clock reference to eliminate distributed clock drift; processing triaxial acceleration data using a low-pass filter with a cutoff frequency adapted to the human motion bandwidth to filter out high-frequency mechanical vibrations; and mapping the high-frequency data from the inertial measurement unit and the low-frequency data from the physiological sensors to a unified target sampling frequency using interpolation or decimation algorithms.

[0009] Preferably, the effective hypoxia work index is constructed as an equivalent physiological cost model weighted by environmental load. This model determines the intensity of physical work by calculating the Euclidean norm of triaxial acceleration and introduces the absolute oxygen partial pressure determined by atmospheric pressure and ambient oxygen concentration as an inverse correction term, reflecting the equivalent physiological load required to generate a unit physical acceleration under a specific altitude oxygen partial pressure environment.

[0010] In one specific embodiment, the cardiovascular response feature is a normalized dimensionless index constructed based on an individual's heart rate reserve. This feature maps the real-time heart rate to a physiological range defined by the historical minimum resting heart rate and the maximum heart rate during maximum exercise load by retrieving the user's individualized physiological baseline parameters, thereby quantifying the actual proportion of the cardiovascular system's utilization relative to its limit capacity.

[0011] In one specific embodiment, the metabolic response feature is a composite feature that integrates static hypoxia depth and dynamic blood oxygen decline rate. It calculates the rate of change of blood oxygen saturation over time by introducing a differential operator, and then uses this rate of change as a negative feedback adjustment term to be superimposed on the current static hypoxia deviation, thereby increasing the output amplitude of the feature value when the absolute value of blood oxygen saturation has not yet reached the alarm threshold but a rapid decline occurs.

[0012] Furthermore, the processing method mapped to the two-dimensional state space is as follows: within the time segment covered by the sliding time window, the effective hypoxia work index is used as the horizontal axis data, and the cardiovascular response characteristics and metabolic response characteristics are used as the vertical axis data, respectively, to construct a dynamic lag trajectory reflecting the physiological regulation relative to the change in physical load, thereby forming a mechanical cardiovascular lag loop and a mechanical metabolic lag loop.

[0013] Preferably, the centroid drift vector is extracted using the discrete geometric moments method. Specifically, the geometric equilibrium position of the hysteresis loop point set within the current window is determined as the dynamic centroid, and a vector is constructed from the pre-stored reference resting-state centroid to the dynamic centroid to determine the direction and distance of the physiological system's deviation from steady state.

[0014] Preferably, the coupling coherence coefficient is configured as a geometric feature that measures the synchronicity of regulation between the cardiovascular and metabolic systems. It is determined by calculating the cosine of the angle between the cardiovascular drift vector and the metabolic drift vector in a two-dimensional state space, and is used to identify whether there is anisotropic decoupling between the two physiological subsystems.

[0015] In one specific embodiment, the comprehensive health risk index is a scalar index that integrates vector direction information, vector magnitude information, and topological area information. This index nonlinearly weights the magnitude of the metabolic drift vector by introducing a decoupling penalty factor that increases with decreasing coupling coherence coefficient, and superimposes a hysteresis loop area term representing physiological energy dissipation.

[0016] Preferably, the tiered early warning decision-making is based on a two-tiered assessment of risk level and decoupling status. The first tier determines whether the comprehensive health risk index exceeds a preset safety threshold; the second tier, when the threshold is exceeded, distinguishes the risk type based on the coupling coherence coefficient: if the coefficient indicates positive synergy, it is determined to be high-load fatigue; if the coefficient indicates anisotropic decoupling, it is determined to be a failure of the compensatory function. This invention provides a method for early warning of health risks among workers in plateau mining industries based on multimodal data. It has the following beneficial effects: 1. This invention constructs an effective hypoxia work index, which inversely weights the physical work intensity represented by triaxial acceleration with the partial pressure of oxygen in the environment. This can quantify the actual physiological cost of the same amount of physical work in hypoxic environments at different altitudes. It solves the problem that single motion data cannot truly reflect the workload of high-altitude work and ensures that the input basis of the risk assessment model can objectively reflect the interaction between the environment and the body.

[0017] 2. This invention utilizes hysteresis loop analysis technology to quantify the dynamic response process of physiological systems by calculating the centroid drift vector of the mechano-cardiovascular and mechano-metabolic hysteresis loops relative to the resting baseline state. Compared to traditional instantaneous physiological index threshold monitoring, this method uses changes in vector direction and magnitude to reflect the degree of deviation from physiological homeostasis, and combines the hysteresis loop area to characterize energy loss, enabling a more accurate assessment of the stability of physiological regulatory functions.

[0018] 3. This invention constructs a coupling coherence coefficient based on the centroid drift vector and distinguishes between positive synergy and anisotropic decoupling states. This mechanism can accurately identify the risk of compensatory failure when the cardiovascular and metabolic systems lose their ability to synchronize regulation. Thus, when the comprehensive health risk index exceeds the standard, it can effectively distinguish between the conventional high-load fatigue state and the pathological compensatory function failure state, thereby improving the specificity of early warning decision-making. Attached Figure Description

[0019] Figure 1 This is a system framework diagram of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention.

[0020] Among them, 10 is a multimodal intelligent terminal; 20 is an environmental monitoring base station; 30 is a data processing platform; 100 is a data acquisition and spatiotemporal alignment module; 200 is a feature construction and mapping module; 300 is a hysteresis dynamics analysis module; and 400 is a risk decision-making module. Detailed Implementation

[0021] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] See attached document Figure 1 The present invention provides a health risk early warning system for plateau mining workers based on multimodal data. The system includes: a multimodal intelligent terminal 10, an environmental monitoring base station 20, and a data processing platform 30.

[0023] The multimodal smart terminal 10 is configured to be worn on specific parts of the body of mining workers for collecting data including real-time heart rate. Blood oxygen saturation and triaxial acceleration , , Physiological and motor data, including those included.

[0024] Environmental monitoring base stations 20 are configured and deployed at the work site to collect data including atmospheric pressure. and ambient oxygen concentration The data processing platform 30 is connected to the multimodal intelligent terminal 10 and the environmental monitoring base station 20 to receive the aforementioned data and perform risk warning calculations.

[0025] The data processing platform 30 includes: a data acquisition and spatiotemporal alignment module 100, a feature construction and mapping module 200, a hysteresis dynamics analysis module 300, and a risk decision-making module 400. These modules execute the data processing flow according to a preset logical sequence, enabling dynamic monitoring and early warning of health risks for workers in plateau mining operations.

[0026] The data acquisition and spatiotemporal alignment module 100 is configured to receive raw data streams from the multimodal intelligent terminal 10 and the environmental monitoring base station 20. The data acquisition and spatiotemporal alignment module 100 uses a timestamp synchronization algorithm to align the multi-source heterogeneous data and follows a unified sampling frequency. Resampling is performed to generate a synchronized time-series dataset containing physiological, kinematic, and environmental signals. Feature construction and mapping module 200 is configured to be based on synchronous time series datasets. We construct a common input basis and dual-channel output response features.

[0027] The feature construction and mapping module 200 first calculates the composite modulus of triaxial acceleration. And the current absolute oxygen partial pressure in the environment ,in: , ; The feature construction and mapping module 200 further calculates the effective hypoxia work index. The calculation formula is: ; In the formula, This is the normalized scaling factor. This index serves as the common input axis data for subsequent hysteresis loop analysis, representing the equivalent physiological load corresponding to a unit physical acceleration under a specific oxygen partial pressure environment.

[0028] The feature construction and mapping module simultaneously computes cardiovascular response features. and metabolic response characteristics Cardiovascular response characteristics The calculation formula is: ; In the formula, and These represent the minimum resting heart rate and the maximum exercise heart rate within a historical period, respectively. Metabolic response characteristics. The calculation formula is: ; In the formula, and These are non-negative weighting coefficients. This represents the rate of change of blood oxygen saturation over time.

[0029] The hysteresis dynamics analysis module 300 is configured to operate at a set length of Within the sliding time window, sets of mechanical cardiovascular hysteresis loop points were constructed respectively. and mechanical metabolic lag loop set . From point pair constitute, From point pair Composition, in which This is the time variable within the window.

[0030] Hysteresis dynamics analysis module 300 calculates the current window. geometric centroid and geometric centroid Subsequently, the hysteresis dynamics analysis module 300, based on the pre-stored reference resting-state centroid... and Construct cardiovascular drift vector and metabolic drift vector The calculation formulas are as follows: The hysteresis dynamics analysis module 300 also uses the discrete Green's formula to calculate the mechanometabolic hysteresis loop. area This area represents the delayed energy loss during the physiological recovery process.

[0031] The risk decision module 400 is configured to perform risk judgments based on the drift vector and the area of ​​the hysteresis loop. The risk decision module 400 calculates the cardiovascular drift vector. With metabolic drift vector Coupling coherence coefficient between : ; In the formula, This represents the Euclidean norm of a vector.

[0032] Risk decision module 400 is based on the coupling coherence coefficient Metabolic drift vector magnitude and the area of ​​the hysteresis loop Calculate the comprehensive health risk index : ; In the formula, To decouple the penalty index, This is the dimensional balance coefficient.

[0033] The risk decision-making module 400 will integrate the health risk index. With preset safety threshold Comparison. When At this time, the risk decision module 400 generates an early warning signal. If at this time... The risk decision module 400 determines that the cardiovascular system and metabolic system have an imbalance in compensatory decoupling and outputs a warning of compensatory function failure.

[0034] See attached document Figure 2 This invention provides a method for early warning of health risks for workers in plateau mining industries based on multimodal data, comprising the following steps: S100 performs spatiotemporal alignment preprocessing on the collected multimodal data and constructs an effective hypoxia work index based on physical work intensity and ambient oxygen partial pressure, which serves as the common input basis for hysteresis loop analysis. S200 extracts physiological response features based on synchronized data, and constructs cardiovascular response features characterizing electromechanical coupling and metabolic response features characterizing mechanochemical coupling, respectively. S300 maps the effective hypoxia work index and physiological response characteristics to a two-dimensional state space, generating a mechanical cardiovascular hysteresis loop and a mechanical metabolic hysteresis loop, and calculates the centroid drift vector of each hysteresis loop respectively. S400 calculates the coupling coherence coefficient based on the centroid drift vector of two hysteresis loops, constructs a comprehensive health risk index by combining the degree of vector anisotropy with the energy loss of the hysteresis loops, and performs graded early warning decisions.

[0035] The specific implementation methods of steps S100 to S400 described above will be described in detail below with reference to specific embodiments.

[0036] The timing synchronization and resampling process of the multi-source data streams in step S100 is specifically executed by the data acquisition and spatiotemporal alignment module 100. The data acquisition and spatiotemporal alignment module 100 first establishes a multi-channel data buffer queue to receive raw data packets from the multimodal intelligent terminal 10 and the environmental monitoring base station 20. Considering the electromagnetic interference and transmission link instability in the mining operation environment, the arrival time of the raw data packets at the data processing platform 30 may be jittery. The data acquisition and spatiotemporal alignment module 100 does not directly use the timestamps generated locally by the sensors. Instead, based on the system atomic time of the data packets arriving at the buffer queue, it reallocates a unified global time stamp for all channels' data streams to eliminate phase errors caused by distributed clock drift.

[0037] After establishing a unified time reference, the data acquisition and spatiotemporal alignment module 100 performs signal conditioning and noise reduction on the raw signals. This is specifically for triaxial acceleration data. , , Because equipment such as rock drills and transport vehicles in mines generate high-frequency mechanical vibrations, these vibrations are considered environmental noise rather than human physical exertion. The data acquisition and spatiotemporal alignment module 100 has an application cutoff frequency of [missing information]. The Butterworth low-pass filter filters acceleration signals, removing high-frequency components with frequencies higher than the bandwidth of normal human motion (e.g., above 20Hz) and retaining low-frequency motion components reflecting human work. The specific design and parameter selection of the Butterworth filter are well-known techniques in the field of digital signal processing and will not be elaborated upon here.

[0038] For physiological signal flow, including heart rate and blood oxygen saturation Because vigorous movement by the wearer may cause momentary loosening of the sensor contact surface, this can introduce spike pulse interference into the signal. The data acquisition and spatiotemporal alignment module 100 employs a moving average filtering algorithm, with a set length of... A sliding window is used to calculate the arithmetic mean of the sampled points within the window as the effective value at the current time, in order to smooth out singularities.

[0039] Even after filtering, the data from each channel still exhibits heterogeneous sampling rate characteristics, with the sampling rate of the inertial measurement unit (IMU) being particularly high. Typically at a much higher sampling rate than heart rate and blood oxygen sensors. To construct the synchronization state vector required for subsequent hysteresis loop analysis, the data acquisition and spatiotemporal alignment module 100 performs a resampling operation, mapping the data from all channels to a unified target sampling frequency. .

[0040] For sampling rates higher than Acceleration data were downsampled using an anti-aliasing decimation method; for sampling rates lower than [a certain value], [the data was further processed]. Physiological and environmental data were used for upsampling and padding using cubic spline interpolation. Cubic spline interpolation constructs a cubic polynomial between adjacent data points. This approximates the true curve and ensures the continuity of the first and second derivatives after interpolation, which is crucial for subsequent calculations of differential characteristics such as the rate of change of blood oxygen.

[0041] The construction of the interpolation function satisfies the following conditions: Let in the interval The interpolation function on is ,but: ; In the formula, , , , The coefficients of the undetermined polynomial are solved using boundary conditions and nodal continuity conditions.

[0042] After completing the above synchronization and resampling, the data acquisition and spatiotemporal alignment module 100 generates a standardized synchronized time series dataset. At any discrete time point ( ), dataset The state matrix is ​​characterized by strict alignment: ; This matrix is ​​passed as standard input to the next-level feature construction and mapping module 200, ensuring that the physiological-load hysteresis loop constructed subsequently has a strict correspondence in the time dimension, and avoiding hysteresis loop morphological distortion caused by phase deviation.

[0043] After the data acquisition and spatiotemporal alignment module 100 completes the data synchronization and standardization process, the feature construction and mapping module 200 uses the synchronized time series dataset as a basis. The effective hypoxia work index is constructed. This step aims to establish a common input base that can span environmental differences across different altitudes, providing a unified load metric for subsequent hysteresis loop analysis.

[0044] The feature construction and mapping module 200 first extracts the triaxial acceleration components from the dataset. By calculating the Euclidean norm, an acceleration modulus capable of characterizing the instantaneous physical intensity of the human body can be synthesized. This modulus eliminates the influence of the direction of motion, focusing only on the total physical work done in an isotropic manner. Its calculation expression is: ; In the formula, These represent the acceleration components along the X, Y, and Z axes of the sensor coordinate system, respectively.

[0045] The feature construction and mapping module 200 simultaneously extracts environmental atmospheric pressure data. and percentage of ambient oxygen concentration In scenarios involving changes in mine ventilation conditions or different working depths, the ambient oxygen partial pressure is dynamically changing. The feature construction and mapping module 200 calculates the absolute oxygen partial pressure of the current working environment based on Dalton's law of partial pressures. This quantifies the degree of oxygen deprivation stress in the environment, and the calculation expression is as follows: ; In the formula, The unit is kilopascal (kPa). These are dimensionless percentage values.

[0046] After obtaining the intensity of physical work and the degree of oxygen deficiency in the environment, the feature construction and mapping module 200 performs nonlinear coupling calculations to generate an effective hypoxia work index. This index is not directly equivalent to physical acceleration; instead, it incorporates environmental oxygen partial pressure as a correction factor to construct an equivalent physiological cost model weighted by environmental load. Its calculation expression is as follows: ; In the formula, This is a preset normalization scaling factor used to map the calculation results to a numerical range that facilitates subsequent processing.

[0047] This computational model introduces The physical significance of the denominator lies in establishing an inverse weighting mechanism. In high-altitude physiology, the same physical power (i.e., the same...) is generated. The required oxygen uptake by the body is negatively correlated with the partial pressure of oxygen in the environment. When the partial pressure of oxygen in the environment... When the load decreases, the human cardiopulmonary system needs to bear a greater internal load in order to maintain the same external work. According to the above formula, at a lower... This will lead to The value increases, thus transforming the external "hard environment" into the internal "soft load" at the numerical level.

[0048] This coupled construction method makes It accurately reflects the equivalent work cost at a specific altitude. As a common horizontal axis (input axis) for subsequently constructing hysteresis loops, it ensures the comparability of miner data operating at different altitudes. For example, the cost of high-intensity labor at low altitudes... The value may be related to the effects of moderate-intensity labor at high altitudes. The values ​​are equal, which accurately reflects that the physiological stress caused by the two is equivalent to that of the body, thus providing a standardized input benchmark for subsequent accurate assessment of physiological compensatory capacity.

[0049] The effective hypoxia work index was completed in the feature construction and mapping module 200. After construction, in step S200, this module performs orthogonal extraction of dual-channel physiological response features in parallel. This process aims to transform a single physiological indicator into dimensionless features capable of characterizing the regulatory state of specific subsystems in the body, namely, cardiovascular dynamic features characterizing the mechanoelectric coupling mechanism. Metabolic response characteristics of characterization coupling mechanisms .

[0050] For the electrokinetic channel, the feature construction and mapping module 200 performs normalization construction based on individual historical extreme values ​​to generate cardiovascular dynamic features. Because mining workers vary in age, basal metabolic rate, physical condition, and altitude acclimatization, using absolute heart rate values ​​directly cannot accurately reflect the actual load on the cardiovascular system. The same heart rate value (e.g., 130 bpm) may only represent a moderate intensity load for a young, physically fit worker, while for an older or weaker worker it may mean approaching their physiological limit.

[0051] To eliminate the influence of individual physical differences on subsequent hysteresis loop morphology analysis, the feature construction and mapping module 200 accesses a pre-set user profile database to retrieve the individualized physiological baseline parameters of the currently wearing employee. These parameters include the employee's minimum resting heart rate over a specific statistical period (such as the past month). and maximum heart rate under maximum exercise load .in, The determination method can be based on the measured peak value in the employee's historical work data, or it can use an estimation formula based on age (such as...). The age can be set.

[0052] The feature construction and mapping module 200, based on the aforementioned benchmark parameters, will use the real-time heart rate data... Mapped to an individual's Heart Rate Reserve (HRR), the calculation formula is as follows: ; In the formula, The current measured heart rate is after filtering. This is the individual's resting baseline heart rate; This refers to the individual's maximum physiological heart rate. This refers to the output cardiovascular dynamic characteristics, and the numerical result is a dimensionless proportionality coefficient.

[0053] Through the above calculations, the feature construction and mapping module 200 converts the absolute heart rate signal with physical units into a normalized index that reflects the relative load intensity. In a physical sense, it represents the proportion of the cardiovascular system's reserves mobilized at the current moment, that is, the actual utilization rate of the heart's pumping function relative to its maximum capacity. When the value approaches 0, it indicates that the body is in a state of complete relaxation; when... When the value approaches 1, it indicates that the body's cardiovascular reserves have been depleted and the body is in a state of extreme work. This normalization process ensures that employees with different physical conditions can exhibit numerically comparable characteristic responses when subjected to the same level of physiological stress, thus providing a standardized ordinate input for subsequently constructing a mechanical cardiovascular hysteresis loop in a unified topological space.

[0054] While constructing cardiovascular dynamics features, the feature construction and mapping module 200 performs metabolic response feature analysis on the mechanotropic channels. The composite construction. Traditional blood oxygen monitoring technology usually only focuses on whether the absolute value of blood oxygen saturation has fallen below a certain safety threshold (e.g., 85%). However, in high-altitude, high-intensity work scenarios, this single-dimensional static monitoring has a lag. When the human body enters a state of acute hypoxia, blood oxygen saturation often undergoes a rapid decline. At this time, even if the absolute value has not yet reached the alarm threshold, the body's gas exchange function is actually in a rapid deterioration phase.

[0055] To overcome this limitation, the feature construction and mapping module 200 introduces a differential operator to construct a composite calculation model that integrates "static hypoxia depth" and "dynamic blood oxygen saturation rate." This model not only assesses the current hypoxic state but also evaluates the trend of hypoxia development. The calculation formula is as follows: ; In the formula, The value is the real-time blood oxygen saturation after preprocessing, ranging from 0 to 100. The item represents the static hypoxia depth, which is the difference between the current blood oxygen level and the full saturation state (100%). The first derivative of blood oxygen saturation with respect to time represents the rate of change in blood oxygen levels. and These are preset non-negative weighting coefficients used to adjust the relative importance of static bias and dynamic change in the feature composition. In discrete sampling systems, the differential term... Through difference operations Implementation, in which The sampling interval is denoted as .

[0056] In this calculation model, the rate of decrease in blood oxygenation is introduced. As a negative feedback term, it has key technical significance. When workers are at a stable blood oxygen level, the derivative term approaches 0, and its characteristic value is mainly determined by the static hypoxia depth. However, in the early stages of acute hypoxia, blood oxygen saturation shows a rapid downward trend, at which point the derivative... The value is negative. This term is multiplied by a negative coefficient in the formula. It was then converted to a positive value and added to the static hypoxia depth term, thereby increasing the The output amplitude.

[0057] This mathematical construction enables metabolic response characteristics. It can keenly capture the rapid changing trends in the early stages of acute hypoxia. Specifically, when the absolute value of blood oxygen is still within a relatively safe range (e.g., 90%) but is rapidly declining, this characteristic value will rise in advance, thus solving the lag problem of traditional threshold monitoring. This feature physically characterizes the degree and trend of decline in the body's gas exchange efficiency, forming the "chemical metabolism" dimension on the vertical axis in hysteresis loop analysis, ensuring that subsequent risk assessment models can cover the entire process from steady-state hypoxia to acute compensatory imbalance.

[0058] The feature construction and mapping module 200 outputs a standardized effective hypoxia work index. Cardiovascular response characteristics and metabolic response characteristics Subsequently, in step S300, the hysteresis dynamics analysis module 300 performs a mapping operation from the one-dimensional time domain to the two-dimensional state space. Because physiological regulatory processes inherently exhibit hysteresis relative to changes in physical load—meaning the body's response path during load increase does not coincide with the recovery path after load removal—simple time-domain waveforms are insufficient to fully describe this dynamic energy dissipation and recovery characteristic. The hysteresis dynamics analysis module 300 captures this dynamic process by establishing a sliding time window mechanism.

[0059] Hysteresis dynamics analysis module 300 sets a length of The time window is defined by the current sampling time. As the cutoff point, backtracking covers a certain time interval. Window length The setting principle covers at least one complete workload cycle or physiological regulation cycle, typically set to 5 to 10 minutes. Within this sliding window, the module extracts the corresponding time series segments and constructs two independent two-dimensional point set trajectories, namely the mechanical cardiovascular hysteresis loop (M-CLoop) and the mechanical metabolic hysteresis loop (M-MLoop).

[0060] For the mechanical cardiovascular hysteresis loop (M-CLoop), the hysteresis dynamics analysis module 300 uses the effective hypoxia work index. The horizontal axis (driving axis) represents the cardiovascular response characteristics. Construct a set of points in a two-dimensional state space, using the ordinate (response axis). Its mathematical expression is: ; In this state space, as time... As time goes by, point set Connecting them sequentially forms a continuous trajectory. In a typical non-steady-state operation, this trajectory appears as a closed or semi-closed loop in a counterclockwise or clockwise direction, and its shape reflects the immediate mobilization capacity and recovery rate of the heart's pumping function in response to the physical-environmental coupled load.

[0061] For the mechanometabolic lag loop (M-MLoop), the lag kinetics analysis module 300 also uses the effective hypoxia work index. The x-axis represents the metabolic response characteristics. Construct another independent two-dimensional point set using the ordinate. Its mathematical expression is: ; The mapping operation here correlates blood oxygen exchange efficiency with physical work done. (Point set) The resulting hysteresis loop characterizes the compensatory pathway of the respiratory metabolic system under hypoxic stress.

[0062] Through the above mapping, the hysteresis dynamics analysis module 300 transforms three signal sequences that were originally independent and difficult to compare directly on the time axis into two geometric figures sharing the same "physical-environmental load" basis (X-axis). This construction method provides a unified topological reference system for subsequent analysis, enabling the assessment of synergistic relationships between different physiological subsystems by comparing the geometrical differences between the two hysteresis loops (such as expansion, contraction, and directional deflection), without relying on direct comparison of absolute values. When new sampling data arrives, the sliding window shifts accordingly, and the hysteresis dynamics analysis module 300 updates the point set in real time. and This enables dynamic tracking of the hysteresis loop morphology.

[0063] After the hysteresis dynamics analysis module 300 generates the hysteresis loop point set trajectory that evolves dynamically over time, in order to quantify this trajectory in two-dimensional state... The module analyzes the overall spatial distribution and migration trends of points. It does not directly handle complex loop-shaped curves, but rather calculates the overall spatial distribution and migration trends of points. Geometric centers are used to extract topological features. The hysteresis dynamics analysis module 300 uses the Discrete Geometric Moment Method to calculate the mechanical cardiovascular hysteresis loops. and mechanical metabolic retardation loop The geometric centroid within the current sliding time window. This calculation process treats all sampling points within the sliding window as particles of equal mass and determines the equilibrium point in the state space by calculating the first moment.

[0064] For mechanical cardiovascular hysteresis loops At the current moment geometric centroid The formula for calculating coordinates is: ; For the mechano-metabolic retardation loop At the current moment geometric centroid The formula for calculating coordinates is: ; In the formula, This represents the total number of sample points within the sliding window. These are variables that iterate through all sampling times within the window. These two centroid coordinates reflect the "load-response" state points of the body's cardiovascular and metabolic systems during the current assessment period.

[0065] To further evaluate the deviation of this average state point from the machine's safety baseline, the hysteresis dynamics analysis module 300 introduces the concept of a "drift vector." The system pre-stores the baseline centroid data of the mining workers in a resting, no-workload state at high altitude, denoted as follows: and These two reference points are typically located near the origin of the coordinate system or in a low-response region, representing the energy balance position of the organism in homeostasis.

[0066] The hysteresis dynamics analysis module 300 constructs a cardiovascular drift vector pointing from the baseline steady state to the current dynamic centroid through vector subtraction. and metabolic drift vector Cardiovascular drift vector The calculation expression is: ; Metabolic drift vector The calculation expression is: ; The drift vector defined above clearly quantifies, in a physical sense, the direction and degree to which a physiological system deviates from its steady state. The vector's magnitude characterizes the absolute strength of physiological compensation, i.e., the extent to which the body expends regulatory effort to cope with the current effective hypoxia index; the vector's direction reflects the "load-response" coupling pattern. For example, when the vector points to the upper right corner (the diagonal direction of the first quadrant), it indicates that the physiological response increases linearly with increasing load, representing normal positive correlation coupling; while an abnormal deflection of the vector direction foreshadows a qualitative change in the regulatory mechanism. This approach of reducing the complex dynamic hysteresis process to vector characteristics provides standardized mathematical input for subsequent accurate identification of decoupling risks between the two systems.

[0067] The hysteresis dynamics analysis module 300 outputs a cardiovascular drift vector that characterizes the dynamic migration state of the cardiovascular and metabolic systems. and metabolic drift vector Subsequently, in step S400, the risk decision module 400 performs deep feature extraction based on vector analysis. The core task of this module is to identify the integrity of the body's compensatory functions by calculating the geometric correlation between the response trajectories of two independent physiological subsystems.

[0068] The risk decision module 400 does not evaluate the magnitude of a single vector in isolation, but focuses on calculating the cosine of the angle between two vectors in the two-dimensional state space, i.e., the coupling coherence coefficient, denoted as . The risk decision-making module 400 calculates based on the vector space cosine similarity theorem. The calculation formula is as follows: ; In the formula, This represents the dot product operation of two vectors. and Let represent the Euclidean norms (i.e., moduli) of the cardiovascular drift vector and the metabolic drift vector, respectively. This is the angle between two vectors. The coefficient... The range of values ​​for is strictly limited to Within the range, its magnitude directly maps the degree of synchronicity in the response of two physiological systems to the same effective hypoxia work index.

[0069] Based on this, the risk decision module 400 executes anisotropic decoupling judgment logic, which is based on the compensatory synergy principle of high-altitude physiology: under normal physiological adaptation or controllable fatigue, as physical load and hypoxia increase, the cardiovascular system compensates by increasing the pumping frequency, while the metabolic system shows a decrease in blood oxygen and an increase in exchange rate. The trends of both are consistent macroscopically, that is, the drift vectors are roughly in the same direction. Approaching 0 degrees, When the value approaches 1, the system determines that it is in an "isotropic cooperative" state.

[0070] However, when the body enters a state of compensatory imbalance or near collapse, physiological regulatory mechanisms exhibit a characteristic "anisotropic" separation. A typical pathological state manifests as a dulling of the cardiovascular system's response, meaning that the heart rate reaches its limit and no longer increases with load, or even reverses course due to abnormally high vagal tone, while the hypoxic condition of the metabolic system continues to worsen. In vector space, this manifests as... Continuing to extend in the original direction, Orthogonal deflection or reverse reversal occurs, resulting in an included angle. Approaching or exceeding 90 degrees. When the risk decision module 400 calculates... Approaching 0 (orthogonal decoupling) or When the value is negative (opposite), the system identifies a break in the direction of cardiovascular regulation and metabolic demand, a state defined as "anisotropic decoupling." This judgment logic effectively distinguishes between normal fatigue under high load (large vector magnitude but consistent direction) and potential risks of sudden death or shock (separated vector direction), thereby achieving precise detection of hidden health risks.

[0071] The risk decision module 400 calculates the coupling coherence coefficient, which reflects the collaborative relationship between systems. Subsequently, this module further combines the absolute deviation of the metabolic dimension with the energy dissipation characteristics of the hysteresis loop to construct a comprehensive health risk index that can fully quantify instantaneous health risks in the human body. This step aims to integrate vector direction information, vector magnitude information, and topological area information into a single scalar index to facilitate standardized threshold decision-making.

[0072] Risk decision module 400 retrieves the area of ​​the mechano-metabolic hysteresis loop calculated by hysteresis dynamics analysis module 300. And the metabolic drift vector magnitude at the current moment. Hysteresis loop area In a physical sense, it characterizes the energy loss or physiological damping caused by the lag in the regulation of the respiratory and metabolic systems during the body's work in response to hypoxia. (The last part, "larger," appears to be an unrelated fragment and is left untranslated.) The numerical value indicates that the body has difficulty recovering to the baseline level quickly after the load is removed, that is, the physiological elasticity is reduced.

[0073] The risk decision-making module 400 calculates the comprehensive health risk index based on the following non-linear weighted formula. : ; In the formula, Represents the absolute magnitude of a metabolic system's deviation from homeostasis; To decouple the penalty factor, The preset decoupling sensitivity index typically takes a value of [value missing]. ; These are the weighting coefficients used to balance the dimensions of the hysteresis energy term.

[0074] In this computational model, This term plays a crucial role in nonlinear amplification. When the cardiovascular and metabolic systems work well together ( When 1), the term approaches 0, even if Larger (indicating high exercise intensity), overall risk index It is also suppressed, reflecting a physiological state of "good adaptation despite heavy load". Conversely, when anisotropic decoupling occurs in the system ( or )hour, Approaching 1 or greater, after exponential After being enlarged, it was improved. The weighting of the system. This means that once an imbalance occurs in the direction of regulation, even if the degree of metabolic deviation has not yet reached an extreme value, the system will calculate an extremely high risk index, thus keenly capturing early signals of compensatory collapse.

[0075] In obtaining a comprehensive health risk index Subsequently, the risk decision-making module 400 executes an adaptive tiered early warning strategy. The system presets a security threshold based on large-scale group data. .

[0076] Risk decision module 400 first judges Does it exceed the threshold? .like If the current condition is determined to be within a safe or controllable compensation range, no alarm will be triggered. The module further relies on the coupling coherence coefficient The numerical characteristics are used to classify and determine the nature of the risk: When detected and This indicates that although the overall risk is high, the cardiovascular and metabolic systems still maintain positive synergy (small angle), and the risk mainly stems from overload or decreased physiological recovery capacity (high). At this point, the risk decision module 400 outputs a "high-load fatigue warning," prompting workers to reduce their workload or take intermittent breaks.

[0077] When detected and When this occurs, it indicates that the system is in a high-risk state accompanied by anisotropic decoupling (vector orthogonal or inverse), meaning a latent compensatory imbalance has occurred. At this point, the risk decision module 400 determines it as a failure of the compensatory function or a latent hypoxic crisis, immediately outputting the highest-level emergency warning and triggering corresponding emergency intervention commands, such as mandatory work stoppage or calling for medical assistance. This tiered mechanism effectively solves the technical problem of traditional methods being unable to distinguish between normal fatigue and pathological collapse.

Claims

1. A high-altitude miner health risk early warning method based on multi-modal data, characterized in that, The method comprises the following steps: S100, physiological data and motion data of plateau mining workers and environmental data of the work site are collected, the collected data are subjected to spatio-temporal alignment preprocessing, and an effective hypoxic work index is constructed based on physical work intensity and environmental oxygen partial pressure as a common input basis for hysteresis loop analysis; S200, based on the spatio-temporal aligned data, double-channel physiological response characteristics are extracted, cardiovascular response characteristics representing force-electric coupling mechanism and metabolic response characteristics representing force-chemical coupling mechanism are constructed; S300, using a sliding time window, the effective hypoxic work index is respectively mapped to a two-dimensional state space with the cardiovascular response characteristics and the metabolic response characteristics, a mechanical cardiovascular hysteresis loop and a mechanical metabolic hysteresis loop are generated, and the centroid drift vectors of each hysteresis loop relative to the resting reference state are calculated; S400, based on the centroid drift vectors of the two hysteresis loops, a coupling coherence coefficient is calculated, a comprehensive health risk index is constructed combining the vector anisotropy degree and the hysteresis loop energy loss, and a graded early warning decision is made according to the comparison result of the comprehensive health risk index and the preset safety threshold.

2. The high-altitude miner health risk early warning method based on multi-modal data according to claim 1, characterized in that, In step S100, the spatio-temporal alignment preprocessing is performed in the following manner: A multi-channel data buffer queue is established to receive the physiological data and motion data including real-time heart rate, blood oxygen saturation, three-axis acceleration, and the environmental data including atmospheric pressure and environmental oxygen concentration; After reassigning global time tags to all data based on a unified system clock reference, low-pass filtering is performed on the three-axis acceleration to retain low-frequency motion components, and anti-aliasing decimation or cubic spline interpolation algorithm is used to resample all data to a unified target sampling frequency.

3. The high-altitude miner health risk early warning method based on multi-modal data according to claim 1, characterized in that, In step S100, the effective hypoxic work index is constructed as an environmental load weighted equivalent physiological cost model, the physical work intensity is determined by calculating the Euclidean norm of the three-axis acceleration, and the absolute oxygen partial pressure determined by atmospheric pressure and environmental oxygen concentration is introduced as a inverse correction factor, which is used to reflect the equivalent physiological load required to produce unit physical acceleration in an altitude oxygen partial pressure environment.

4. The high-altitude miner health risk early warning method based on multi-modal data according to claim 1, characterized in that, In step S200, the cardiovascular response characteristics are normalized dimensionless indexes constructed based on individual heart rate reserve intervals, the real-time heart rate is mapped to the physiological interval defined by the historical resting heart rate minimum value and the maximum exercise load heart rate maximum value by calling the user's individual physiological reference parameters, and the actual mobilization proportion of the cardiovascular system relative to the user's limit capacity is represented.

5. The high-altitude miner health risk early warning method based on multi-modal data according to claim 1, characterized in that, In step S200, the metabolic response characteristics are a composite feature combining static hypoxia depth and dynamic blood oxygen drop rate, the time change rate of blood oxygen saturation is calculated by introducing a differential operator, and the change rate is added as a negative feedback item to the current static hypoxia deviation, so as to increase the output amplitude of the feature value in advance when the absolute value of blood oxygen saturation has not reached the alarm threshold but has a rapid drop.

6. The high-altitude miner health risk early warning method based on multi-modal data according to claim 1, characterized in that, In step S300, the mapping to the two-dimensional state space is processed in the following manner: In the time sequence segment covered by the sliding time window, the effective hypoxic work index is taken as driving shaft data, and the cardiovascular response feature and the metabolic response feature are taken as response shaft data, a dynamic hysteresis trajectory reflecting the change of physiological regulation relative to physical load is constructed, thereby forming the mechanical cardiovascular hysteresis loop and the mechanical metabolic hysteresis loop.

7. The high-altitude miner health risk early warning method based on multi-modal data according to claim 6, characterized in that, In step S300, the centroid drift vector is extracted by discrete geometric moment method, specifically, the geometric balance position of the hysteresis loop point set in the current window is determined as a dynamic centroid, and a vector from the pre-stored reference resting state centroid to the dynamic centroid is constructed.

8. The high-altitude miner health risk early warning method based on multi-modal data according to claim 1, characterized in that, In step S400, the coupling coherence coefficient is configured to measure the geometric feature of the synchronization of cardiovascular system and metabolic system regulation, is determined by calculating the cosine value of the included angle between the cardiovascular drift vector and the metabolic drift vector in the two-dimensional state space, and is used to identify whether there is anisotropic decoupling between the two physiological subsystems.

9. The high-altitude miner health risk early warning method based on multi-modal data according to claim 1, characterized in that, In step S400, the comprehensive health risk index is a scalar index that integrates vector direction information, vector length information and topological area information, the length of the metabolic drift vector is nonlinearly weighted by introducing a decoupling penalty factor that increases as the coupling coherence coefficient decreases, and a hysteresis loop area term representing physiological energy dissipation is superimposed.

10. The high-altitude miner health risk early warning method based on multi-modal data according to claim 1, characterized in that, In step S400, the hierarchical early warning decision is based on the risk degree and the decoupling state for double-layer judgment: The first layer judges whether the comprehensive health risk index exceeds the preset safety threshold; The second layer, when exceeding the preset safety threshold, distinguishes the risk type according to the coupling coherence coefficient, and if the coupling coherence coefficient indicates positive cooperation, it is judged as high-load fatigue, and if the coupling coherence coefficient indicates anisotropic decoupling, it is judged as compensatory function failure.

Images