Plateau energy high-risk scene personnel vital signs and environment collaborative monitoring and early warning method

Abstract

The invention discloses a method for plateau energy high-risk scene personnel vital signs and environment collaborative monitoring and early warning, comprising: projecting the time series data of air pressure, oxygen partial pressure and temperature and humidity into the perturbation tensor space, extracting the environmental stress gradient, and based on the gradient, implementing joint probability density reconstruction on multi-modal physiological parameters such as blood oxygen saturation, heart rate and respiratory rate to generate a sign distortion indicator. The probability distribution of the indicator is aligned in the hidden space by using the variational autoencoder, the abnormal judgment threshold is dynamically calculated, then the sliding window integral wandering test is performed on the threshold sequence, and the covariant mutation point between the environment and the sign is captured. According to the space-time coordinates of the mutation point, the confrontation game constraint of the environmental stress weight and the sign distortion weight is embedded, and the output of the judgment includes the collaborative warning level signal of the environmental risk intensity component and the sign instability intensity component, which improves the adaptive ability and reliability of the early warning system in the plateau energy high-risk scene.

Description

Technical Field

[0001] This invention relates to the field of monitoring and early warning technology for high-risk operation scenarios, and in particular to a method for coordinated monitoring and early warning of personnel vital signs and the environment in high-risk energy scenarios on plateaus. Background Technology

[0002] The high-altitude energy operation environment is characterized by extreme features such as sudden drops in air pressure, sharp decreases in oxygen partial pressure, drastic temperature variations, and rapid climate change. In such high-risk scenarios, there is a strong, nonlinear, dynamic interaction between the physiological state of workers and environmental stressors. With the continuous expansion of high-altitude mineral resource development and energy infrastructure construction, the frequency and duration of workers' exposure to multiple stressors such as hypoxia, low temperature, and strong radiation have increased significantly. The risk of sudden altitude sickness and operational errors has risen sharply, posing a serious threat to personnel safety and the continuity of production operations. Therefore, how to implement coordinated monitoring of environmental disturbances and personnel vital signs and achieve early warning has become a key technical problem that urgently needs to be solved in the field of high-altitude energy safety production.

[0003] Existing technologies typically employ a discrete monitoring architecture, deploying environmental parameter acquisition systems and personnel physiological parameter acquisition systems independently. On the environmental side, time-series environmental data is acquired through barometric pressure sensors, oxygen partial pressure sensors, and temperature and humidity sensors. On the personnel side, multimodal physiological parameters are acquired through pulse oximeters, heart rate monitors, and respiratory monitoring devices. Data from both sides are processed by their respective independent threshold judgment logic to generate alarm signals. Some improved solutions introduce data fusion strategies, weighting and superimposing environmental and physiological parameters at the decision-making level. Early warnings are triggered based on preset fixed thresholds, and time-series trend analysis is used to track and identify parameter drift, thereby providing alerts for abnormal states exceeding the normal fluctuation range.

[0004] However, existing technologies of this kind have the following drawbacks. First, the coupling perturbation relationship between environmental and physiological parameters is not effectively modeled. Existing methods treat environmental changes as exogenous variables and vital sign changes as independent responses, neglecting the dynamic driving mechanism of environmental stress on vital sign distortion and the covariant transmission effect between the two over time. This results in delayed early warning triggering and a persistently high false alarm rate. Second, the anomaly judgment threshold adopts a static setting strategy, which cannot adaptively track the baseline drift of vital signs caused by non-stationary perturbations in high-altitude environments. Fixed thresholds frequently trigger false alarms during periods of drastic environmental fluctuations, while severely missing alarms during the accumulation phase of slowly changing stresses. This makes it difficult to meet the reliability requirements of collaborative monitoring and early warning in high-risk scenarios. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of existing technologies, this invention provides a method for coordinated monitoring and early warning of personnel vital signs and the environment in high-risk energy scenarios on plateaus.

[0006] The technical solution adopted in this invention is a method for coordinated monitoring and early warning of personnel vital signs and the environment in high-risk energy scenarios on plateaus, including the following steps: S1, constructing a plateau environmental disturbance feature mapping algorithm, projecting the collected air pressure, oxygen partial pressure, and temperature and humidity time series data into a disturbance tensor space, and extracting the environmental stress gradient; S2, based on the output of the disturbance tensor space, starting a high-altitude hypoxia vital sign distortion identification model, performing joint probability density reconstruction on multimodal physiological parameters such as blood oxygen saturation, heart rate, and respiratory rate, and generating vital sign distortion indicator factors; S3, transferring the distortion indicator factors... The input variational autoencoder anomaly threshold coupler is used to calculate the dynamic anomaly decision threshold through latent space probability distribution alignment operation; S4, the non-stationary temporal mutation detection and analysis platform is used to perform a sliding window integral walk test on the threshold sequence to capture the covariant mutation point between environmental disturbance and physical sign drift; S5, based on the spatiotemporal coordinates of the mutation point, the collaborative early warning decision logic is activated, and the decision logic is embedded with the adversarial game constraint of environmental stress weight and physical sign distortion weight; S6, the collaborative monitoring early warning level signal is output, which includes the environmental hazard intensity component and the physical sign instability intensity component.

[0007] Furthermore, the expression for the plateau environment disturbance feature mapping algorithm is as follows: ;in, It represents the cumulative potential of environmental disturbances. For the spatial integration field, It is a spatial position vector. The pressure field distribution function is... The enthalpy-humidity coupling potential function is... For time variables, This is the pressure gradient weighting coefficient. The enthalpy-humidity time-varying coupling coefficient is... For perturbation mapping quantization value, The total number of sampling points. For sampling point index, It is a nonlinear modulation factor. It is the stability constant. For the first The spatial weighted index of each sampling point, For the first The cumulative potential of the perturbation at each sampling point is taken as follows: For gradient operators, This is the symbol for partial differentials.

[0008] Furthermore, the expression for the high-altitude hypoxia sign distortion recognition model is as follows: ;in, For multimodal physiological parameter joint bias functional, This represents the total number of physiological parameter categories. For category indexing, For the first Deviation weights of class parameters For the first Real-time observations of class parameters For the first The baseline center value of the class parameter, For the first The baseline scaling factor for class parameters For the first The saturation suppression coefficient of the class parameter, For the first The sliding difference value of the class parameter, for Differential scaling constants for class parameters, The cumulative response for distortion identification. This is the upper limit of the integration time window. For integration time variable, for The joint bias functional values ​​at time t, It is the saturation cutoff constant. for The decay function of ambient oxygen partial pressure at time t. This is the differential symbol.

[0009] Furthermore, the expression for the variational autoencoder anomaly threshold coupler is: ,in, This is the threshold value for dynamic anomaly detection. The total number of dimensions encoded in the latent space. For dimensional indexing, For the first Dimensional contribution scaling factor For the variational autoencoder to encode the distortion indicator factor, at the 1st... The values ​​of the latent variables in the dimension, For the first The baseline cluster center vector of dimension, For the Euclidean norm, For the first The radial bandwidth coefficient of the dimension, For the first Multiplicative exponent weights of dimensions.

[0010] Furthermore, in the sliding window integral walk test, the sliding window length is 60 sampling points, the window overlap rate is 50%, and the walk step size of the integral walk test is 5 sampling points; in the adversarial game constraint, the environmental stress weight is 0.55 to 0.65, the physical characteristic distortion weight is 0.35 to 0.45, and the sum of the two weights is always 1; the collaborative early warning level signal is divided into four levels, and the level division is based on the comparison between the weighted product of the environmental risk intensity component and the physical characteristic instability intensity component and the decision threshold; the decision threshold includes a high threshold value and a low threshold value, the high threshold value is 1.5 times the dynamic anomaly decision threshold, and the low threshold value is 0.6 times the dynamic anomaly decision threshold.

[0011] Further, the specific steps of the joint probability density reconstruction performed by the high-altitude hypoxia physical sign distortion identification model in S2 are as follows: S21, the multimodal physiological parameters of blood oxygen saturation, heart rate and respiratory rate are respectively truncated in the time domain, and the truncated length is aligned with the output timestamp of the perturbation tensor space; S22, based on the truncated parameter segments, the empirical cumulative distribution function value of each parameter in the current segment is calculated; S23, the empirical cumulative distribution function value of each parameter is projected to the high-dimensional joint probability density space, and a joint distribution surface is generated by using a kernel density estimation strategy; S24, the probability density offset of the joint distribution surface relative to the reference distribution surface is extracted, and the offset is accumulated to obtain the physical sign distortion indicator factor.

[0012] Further, the specific steps of the variational autoencoder anomaly threshold coupler performing latent space probability distribution alignment in S3 are as follows: S31, encoding the symptom distortion indicator factor into a mean and variance parameter pair in the latent space, and outputting the parameter pair by the encoding network of the variational autoencoder; S32, resampling latent variable samples from the Gaussian distribution described by the mean and variance parameter pair; S33, calculating the KL divergence value between the latent variable sample and the pre-stored normal latent space distribution; S34, inputting the KL divergence value into the threshold mapping function, and outputting the dynamic anomaly decision threshold through the nonlinear transformation of the threshold mapping function.

[0013] Furthermore, the specific steps of the non-stationary temporal mutation detection and analysis platform in S4 for performing the sliding window integral walk test on the threshold sequence are as follows: S41, set a sliding window on the threshold sequence, and calculate the local integral area of ​​the threshold sequence in each window; S42, recursively push the local integral area forward according to the walk step size to generate the integral area walk trajectory; S43, perform a signed-rank test on the walk trajectory to obtain the deviation rank sum value of the integral area in each window from the median of the global integral area; S44, mark the window with the deviation rank sum value exceeding the preset walk threshold as the covariant mutation point window, and extract the spatiotemporal coordinates of the mutation point.

[0014] Furthermore, the specific steps of embedding the environmental stress weight and physical characteristic distortion weight adversarial game constraint in the collaborative early warning adjudication logic in S5 are as follows: S51, extract the current environmental stress intensity value and the current physical characteristic distortion intensity value according to the spatiotemporal coordinates of the mutation point; S52, multiply the environmental stress intensity value and the physical characteristic distortion intensity value by the environmental stress weight and the physical characteristic distortion weight respectively to obtain the weighted stress value and the weighted distortion value; S53, calculate the difference between the weighted stress value and the weighted distortion value, and use this difference as the game adversarial deviation; S54, based on the numerical range of the game adversarial deviation, adjudicate and output the corresponding early warning level signal.

[0015] A collaborative monitoring and early warning method for personnel vital signs and the environment in high-risk energy scenarios on plateaus is proposed. This method is implemented through a collaborative monitoring and early warning system for personnel vital signs and the environment in high-risk energy scenarios on plateaus, comprising: a plateau environmental disturbance feature mapping module, a high-altitude hypoxia vital sign distortion identification module, a variational autoencoder anomaly threshold coupling module, a non-stationary temporal mutation detection and analysis module, a collaborative early warning adjudication game module, and an early warning level output module. The plateau environmental disturbance feature mapping module receives externally acquired temporal data of air pressure, oxygen partial pressure, and temperature and humidity, and outputs a disturbance tensor space to the high-altitude hypoxia vital sign distortion identification module. The high-altitude hypoxia vital sign distortion identification module receives the output of the disturbance tensor space and integrates externally acquired multimodal physiological parameters such as blood oxygen saturation, heart rate, and respiratory rate. The system outputs a vital sign distortion indicator factor to the variational autoencoder anomaly threshold coupling module; the variational autoencoder anomaly threshold coupling module performs latent space probability distribution alignment on the vital sign distortion indicator factor and outputs a dynamic anomaly judgment threshold to the non-stationary temporal mutation detection and analysis module; the non-stationary temporal mutation detection and analysis module performs a sliding window integral walk test on the threshold sequence and outputs the spatiotemporal coordinates of the covariant mutation point to the collaborative early warning adjudication game module; the collaborative early warning adjudication game module embeds an adversarial game constraint of environmental stress weight and vital sign distortion weight based on the spatiotemporal coordinates of the mutation point and outputs an early warning level signal to the early warning level output module; the early warning level output module decomposes the early warning level signal into an environmental hazard intensity component and a vital sign instability intensity component and presents them externally.

[0016] Beneficial Effects: This invention proposes a collaborative monitoring and early warning method for personnel vital signs and the environment in high-risk energy scenarios on plateaus. The constructed environmental disturbance feature mapping algorithm can project multi-source environmental parameters such as air pressure, oxygen partial pressure, temperature, and humidity onto a unified disturbance tensor space, characterizing the spatiotemporal evolution gradient of plateau environmental stress. This provides a dynamic environmental benchmark for identifying vital sign distortions, overcoming the deficiency in existing technologies that treat environmental changes as exogenous variables and ignore their driving role. The high-altitude hypoxia vital sign distortion identification model captures the nonlinear distortion characteristics of vital sign responses driven by environmental disturbances by performing joint probability density reconstruction on multimodal physiological parameters such as blood oxygen saturation, heart rate, and respiratory rate. This allows for the quantitative expression of the covariant transmission effect between the environment and vital signs, solving the problem of early warning lag caused by separate monitoring of the environment and vital signs. The variational autoencoder anomaly threshold coupler dynamically calculates the anomaly decision threshold by aligning the latent space probability distribution. This allows the threshold value to adaptively adjust with the degree of environmental disturbance and baseline drift of vital signs, breaking free from the constraints of static fixed thresholds. This reduces the false alarm rate during periods of severe fluctuations and eliminates the risk of missed alarms during the gradual accumulation phase. The non-stationary temporal abrupt change detection and analysis platform, by performing a sliding window integral walk test on the threshold sequence, can accurately locate the covariant abrupt change point between environmental disturbances and vital sign drift, providing accurate spatiotemporal trigger coordinates for early warning decisions. The adversarial game constraint of environmental stress weights and vital sign distortion weights embedded in the collaborative early warning decision logic ensures that the early warning level judgment takes into account the dynamic game relationship between the intensity of environmental hazards and the severity of vital sign instability. The output signal includes dual components, which can guide graded response measures. This method realizes the collaborative monitoring of environmental and physiological parameters throughout the entire process from data acquisition and feature fusion to decision output, improving the adaptability and reliability of the early warning system in high-risk scenarios of plateau energy, and solving the shortcomings of existing technologies such as high false alarm rates, serious missed alarms, and lack of dynamic adjustment capability of decision thresholds. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention. Figure 2 This is a flowchart of method step S2 of the present invention; Figure 3 This is a flowchart of method step S3 of the present invention; Figure 4 This is a flowchart of method step S4 of the present invention; Figure 5 This is a flowchart of step S5 of the method of the present invention. Detailed Implementation

[0018] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. The application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] like Figure 1 As shown, a collaborative monitoring and early warning method for personnel vital signs and the environment in high-risk energy scenarios on plateaus includes the following steps: S1, constructing a plateau environmental disturbance feature mapping algorithm, projecting the collected air pressure, oxygen partial pressure, and temperature and humidity time-series data onto the disturbance tensor space, and extracting the environmental stress gradient; S2, based on the output of the disturbance tensor space, activating the high-altitude hypoxia vital sign distortion identification model, performing joint probability density reconstruction on multimodal physiological parameters such as blood oxygen saturation, heart rate, and respiratory rate, and generating vital sign distortion indicator factors; S3, inputting the distortion indicator factors into the variational autogenous system... The system encodes an anomaly threshold coupler and calculates the dynamic anomaly decision threshold through latent space probability distribution alignment. In step S4, a sliding window integral walk test is performed on the threshold sequence using a non-stationary temporal mutation detection and analysis platform to capture covariant mutation points between environmental disturbances and physical sign drift. In step S5, based on the spatiotemporal coordinates of the mutation points, the collaborative early warning decision logic is activated, embedding an adversarial game constraint between environmental stress weights and physical sign distortion weights. In step S6, a collaborative monitoring early warning level signal is output, which includes environmental hazard intensity components and physical sign instability intensity components.

[0020] In step S1, a distributed environmental sensing node array is deployed in the plateau energy operation area. Each node integrates a pressure sensing unit, an oxygen partial pressure sensing unit, and a temperature and humidity composite sensing unit. Each sensing unit synchronously collects four time-series signals—pressure, oxygen partial pressure, temperature, and relative humidity—at a sampling frequency of 10 Hz. The collected raw signals are converted from analog to digital to form a digital sequence. This sequence is then aligned according to the data acquisition timestamps and input into a perturbation feature mapping processor. The processor internally constructs a six-dimensional perturbation tensor space. The six dimensions correspond to the rate of change of pressure, the attenuation gradient of oxygen partial pressure, the amplitude of temperature fluctuation, the coefficient of variation of humidity, the covariance coefficient of pressure and oxygen partial pressure, and the temperature-humidity coupling index. The processor performs sliding differential operations on each signal, with a differential window length of 50 sampling points and a sliding step size of 10 sampling points, calculating the values ​​of each tensor component in each dimension window by window. The calculated six-dimensional tensor components are arranged in chronological order to form a perturbation tensor time series. Spatial interpolation is then performed on this series with an interpolation grid resolution of 0.5 meters. After interpolation, the environmental stress gradient vector at each grid node is extracted. This vector is composed of the spatial partial derivatives of the tensor components, and the magnitude of the stress gradient vector characterizes the overall intensity level of the environmental perturbation at the current moment. Finally, the stress gradient vectors of all grid nodes are encapsulated into a perturbation tensor spatial output.

[0021] In step S2, a wearable physiological monitoring patch is attached to the worker's body surface. This patch integrates a reflective blood oxygen sensor, a piezoelectric heart rate sensor, and an impedance-type respiratory sensor band. The three sensors simultaneously collect three multimodal physiological parameters—blood oxygen saturation, heart rate, and respiratory rate—at a sampling frequency of 25 Hz. The collected time-series physiological parameter data is strictly aligned with the perturbation tensor space data output from S1 according to timestamps. The aligned data is then input into a vital sign distortion recognition processor. The processor first performs time window segmentation on the three types of physiological parameters, with a time window width set to 120 sampling points and an overlap rate of 25% between adjacent time windows. Within each time window, the decrease slope of blood oxygen saturation, the variance of heart rate fluctuation, and the amplitude of respiratory rate variation are calculated. These three calculated values ​​are combined into a three-dimensional feature vector, which is then paired with the environmental stress gradient vector in the perturbation tensor space at corresponding times. After pairing, the magnitude of the environmental stress gradient vector is used as a conditional variable to perform conditional probability density mapping on the three-dimensional feature vector. The mapping process uses a multivariate Gaussian kernel function for density estimation, with the diagonal elements of the kernel function bandwidth matrix taking values ​​of 2% for blood oxygen saturation, 5 beats per minute for heart rate, and 3 breaths per minute for respiratory rate. The probability density value of the feature vector at the current moment on the conditional probability density surface is calculated, and the absolute difference between the probability density value under the baseline state and the current probability density value is output as the raw numerical value of the vital sign distortion indicator.

[0022] In step S3, the variational autoencoder anomaly threshold coupler receives the time-series sequence of vital sign distortion indicators output from S2, which is continuously input at an update rate of 10 Hz. The coupler internally deploys a variational autoencoder network structure, comprising an input layer, hidden layers, and an output layer. The number of neurons in the hidden layer is set to 64, and the activation function is a nonlinear saturation function. The encoder maps the currently input vital sign distortion indicators to mean and variance parameters in the latent space, which is set to 8 dimensions. A Gaussian distribution is constructed based on the mean and variance parameters, and a random resampling operation is performed from this Gaussian distribution to obtain 8-dimensional latent variable samples. The coupler internally pre-stores the baseline distribution parameters of the latent space collected under normal operating conditions, including the baseline mean vector and the baseline covariance matrix. The divergence value of the probability distribution between the current latent variable sample and the baseline distribution is calculated using an integral probability metric. The calculated divergence value is input into a threshold mapping network, which is a single-layer linear transformation structure with an 8x1 matrix and a fixed offset term. The output value of the threshold mapping network is then numerically clamped, with a lower clamp limit of 0.05 and an upper clamp limit of 2.0. The clamped output value serves as the dynamic anomaly detection threshold for the current moment, and this threshold value is updated sequentially with each newly arrived vital sign distortion indicator.

[0023] In step S4, the non-stationary temporal mutation detection and analysis platform receives the time series of dynamic anomaly judgment thresholds from S3, with the data sampling rate of the threshold sequence consistent with the output of S3. An internal circular buffer is set up to store historical values ​​of the threshold sequence, with a buffer capacity of 600 sampling points. A sliding window is defined on the circular buffer, with a fixed width of 60 sampling points and a step size of 5 sampling points for each forward slide. The overlap between adjacent windows is 55 sampling points. For each sliding window position, the deviation sequence of all 60 threshold values ​​within the window relative to the global threshold mean is calculated. Numerical integration is performed on this deviation sequence using the trapezoidal rule, with the integration step size between adjacent sampling points equal to the sampling interval. The calculated cumulative value is used as the local integration area of ​​that window. The local integration areas are arranged according to the time sequence of the window sliding to form an integration area walking trajectory sequence. A signed-rank statistical test is performed on the walking trajectory sequence, with the null hypothesis that the local integration area randomly fluctuates around the median of the global integration area. The rank statistic for each window is calculated. Windows whose absolute value of the rank statistic exceeds the critical value of 5.0 are identified as deviation windows. The earliest sampling point with the time index in the deviation window is extracted as the time coordinate of the covariant mutation point. At the same time, the perturbation tensor space grid coordinates corresponding to the mutation point are extracted as spatial coordinates. The combination forms the spatiotemporal coordinate output of the mutation point.

[0024] In step S5, the collaborative early warning adjudication module receives the spatiotemporal coordinates of the mutation point from S4, extracts the magnitude of the environmental stress gradient vector corresponding to the spatiotemporal coordinates from the perturbation tensor space of S1 as the environmental stress intensity value, and extracts the physical characteristic distortion indicator factor value at the same moment from the physical characteristic distortion identification model of S2 as the physical characteristic distortion intensity value. The adjudication module internally configures environmental stress weight factors and physical characteristic distortion weight factors, with initial weight factors set to 0.60 and 0.40 respectively. The environmental stress intensity value is multiplied by the environmental stress weight factor to obtain the weighted stress value, and the physical characteristic distortion intensity value is multiplied by the physical characteristic distortion weight factor to obtain the weighted distortion value. The difference between the weighted stress value and the weighted distortion value is calculated as the game adversarial bias. Based on the sign and absolute value of the deviation, an online adjustment operation of the weighting factors is performed. If the deviation is positive and the absolute value is greater than 0.1, the weighting factor is adjusted in increments of 0.01 towards increasing the weighting factor for physical deformity, while the weighting factor for environmental stress is correspondingly decreased. If the deviation is negative and the absolute value is greater than 0.1, the weighting factor is adjusted in increments of 0.01 towards increasing the weighting factor for environmental stress, while the weighting factor for physical deformity is correspondingly decreased. Throughout the adjustment process, the sum of the two weighting factors is always kept equal to 1.0. The adjusted weighting factors are used for the next deviation calculation, forming a cyclical iterative adversarial game process.

[0025] When step S6 is implemented, the early warning signal output module receives the weighted coercion value and weighted distortion value after each game adjustment from S5, and simultaneously obtains the dynamic anomaly judgment threshold at the current moment from S3. The weighted coercion value and the weighted distortion value are added together to obtain the comprehensive risk value. The module internally stores the trigger threshold ranges corresponding to four early warning levels: the comprehensive risk value range for level one is greater than or equal to 0 and less than 0.2; the comprehensive risk value range for level two is greater than or equal to 0.2 and less than 0.5; the comprehensive risk value range for level three is greater than or equal to 0.5 and less than 0.8; and the comprehensive risk value range for level four is greater than or equal to 0.8. The comprehensive risk value is compared with the dynamic anomaly judgment threshold. If the comprehensive risk value is less than 0.6 times the dynamic anomaly judgment threshold, a Level 1 signal is output. If the comprehensive risk value is greater than or equal to 0.6 times the dynamic anomaly judgment threshold but less than the dynamic anomaly judgment threshold, a Level 2 signal is output. If the comprehensive risk value is greater than or equal to the dynamic anomaly judgment threshold but less than 1.5 times the dynamic anomaly judgment threshold, a Level 3 signal is output. If the comprehensive risk value is greater than or equal to 1.5 times the dynamic anomaly judgment threshold, a Level 4 signal is output. The final warning level signal is output in digital code form, using a two-bit binary number, corresponding to the environmental hazard intensity component and the physical instability intensity component, respectively. The two components are independently encoded and then combined to form a complete warning level digital code for output.

[0026] Preferably, the expression for the plateau environment disturbance feature mapping algorithm is: ;in, It represents the cumulative potential of environmental disturbances. For the spatial integration field, It is a spatial position vector. The pressure field distribution function is... The enthalpy-humidity coupling potential function is... For time variables, This is the pressure gradient weighting coefficient. The enthalpy-humidity time-varying coupling coefficient is... For perturbation mapping quantization value, The total number of sampling points. For sampling point index, It is a nonlinear modulation factor. It is the stability constant. For the first The spatial weighted index of each sampling point, For the first The cumulative potential of the perturbation at each sampling point is taken as follows: For gradient operators, This is the symbol for partial differentials.

[0027] Specifically, the plateau environment disturbance feature mapping algorithm constructs the cumulative potential of environmental disturbance, based on the principle of energy conservation in the extended form of the non-isothermal atmospheric boundary layer. The gradient of the pressure field distribution function in space characterizes the horizontal pressure gradient force, which drives atmospheric motion and directly affects the oxygen partial pressure transport efficiency; the partial derivative of the enthalpy-humidity coupling potential function with respect to time characterizes the latent heat flux disturbance induced by the joint changes in temperature and absolute humidity. This disturbance has different spatiotemporal scales than the pressure gradient disturbance, but both act on the human respiratory and circulatory systems. Multiplying both by the pressure gradient weighting coefficient and the enthalpy-humidity time-varying coupling coefficient respectively, and then performing a volume integral in the spatial integration domain, the physical meaning is the total spatial accumulation of environmental disturbance energy in the entire operating area. The weighting coefficient is set according to the altitude classification of the plateau: 0.42 for altitudes of 3000-3500 meters, 0.55 for altitudes of 3500-4000 meters, and 0.68 for altitudes above 4000 meters. The coupling coefficient is dynamically adjusted within the range of 0.15 to 0.45 according to the diurnal temperature difference. A disturbance mapping quantization value is established based on the nonlinear normalization processing of the accumulated potential at discrete sampling points. A multiplication operation is used instead of a summation operation because the disturbance effects at each sampling point have a product coupling relationship; an extreme disturbance at one sampling point will amplify the disturbance contribution of adjacent sampling points. The nonlinear modulation factor is set to 0.75, and the stability constant is set to 0.01 to prevent the denominator from being too small. The spatial weighting index is allocated according to the radial distance between the sampling point and the worker's activity trajectory. When the distance is less than 5 meters, it is set to 1.0; when the distance is between 5 and 20 meters, it is set to 0.6; and when the distance is greater than 20 meters, it is set to 0.3.

[0028] Preferably, the expression for the high-altitude hypoxia physical sign distortion recognition model is: ;in, For multimodal physiological parameter joint bias functional, This represents the total number of physiological parameter categories. For category indexing, For the first Deviation weights of class parameters For the first Real-time observations of class parameters For the first The baseline center value of the class parameter, For the first The baseline scaling factor for class parameters For the first The saturation suppression coefficient of the class parameter, For the first The sliding difference value of the class parameter, For the first Differential scaling constants for class parameters, The cumulative response for distortion identification. This is the upper limit of the integration time window. For integration time variable, for The joint bias functional values ​​at time t, It is the saturation cutoff constant. for The decay function of ambient oxygen partial pressure at time t. This is the differential symbol.

[0029] Specifically, in the high-altitude hypoxia sign distortion identification model, a multimodal physiological parameter joint bias functional is constructed based on an improved framework using Mahalanobis distance. For the three physiological parameters of blood oxygen saturation, heart rate, and respiratory rate, the difference between the real-time observed value and the baseline center value of each parameter is divided by the baseline scaling factor to obtain the standardized bias. The square of this bias multiplied by the bias weight represents the cost loss of the parameter deviating from the normal state. In addition, the sliding difference value of the real-time observed value is scaled by the saturation inhibition coefficient and then the logarithm is taken. This term is used to capture the rate information of changes in signs. Since slow drift and rapid change have different meanings for early warning decisions, they need to be modeled separately. The bias weights were set based on the sensitivity of each parameter to changes in ambient oxygen partial pressure: blood oxygen saturation was set to 1.0, heart rate to 0.6, and respiratory rate to 0.4. The baseline center value was obtained by taking the median of 120 seconds of physiological data collected from workers at rest. The baseline scaling factor was calculated by dividing the interquartile range of the 120-second data by 1.35. The saturation suppression coefficient was used to adjust the degree of nonlinear compression of the difference value, with values ​​of 2.5, 1.8, and 2.0 for the three types of parameters, respectively. The cumulative response value for distortion identification, calculated by integrating it over time, is based on the medical and physiological principle of accumulating the instantaneous bias functional over exposure time. The cumulative exposure value is a better indicator of the progressive characteristics of hypoxia injury than the instantaneous value. The upper limit of the integration time window is set to 180 seconds, which matches the typical time constant of the human circulatory system's response to hypoxia stress. The nonlinear term of the ratio of the deviation functional to the oxygen partial pressure decay function in the integral reflects the physical fact that a lower oxygen partial pressure produces a stronger physiological impact under the same deviation. The saturation cutoff constant is set to 0.005 to prevent the numerical explosion caused by an excessively small denominator.

[0030] Preferably, the expression for the variational autoencoder anomaly threshold coupler is: ,in, This is the threshold value for dynamic anomaly detection. The total number of dimensions encoded in the latent space. For dimensional indexing, For the first Dimensional contribution scaling factor For the variational autoencoder to encode the distortion indicator factor, at the 1st... The values ​​of the latent variables in the dimension, For the first The baseline cluster center vector of dimension, For the Euclidean norm, For the first The radial bandwidth coefficient of the dimension, For the first Multiplicative exponent weights of dimensions.

[0031] Specifically, the variational autoencoder anomaly threshold coupler transforms the radial distance of the encoded vectors in the latent space into probabilistic membership degrees, and then obtains the threshold value through nonlinear weighted multiplication. After the encoding network of the variational autoencoder reduces the dimensionality of the vital sign distortion indicator to the latent space, the smaller the Euclidean distance between the latent variable value in each dimension and the pre-stored benchmark cluster center vector, the closer the current vital sign state is to the normal state; the larger the distance, the more it deviates from the normal state. For all eight dimensions of the latent space, the Gaussian radial basis function value is calculated independently for each dimension. The exponent term of the radial basis function is the square of the negative Euclidean distance divided by the radial bandwidth coefficient. The bandwidth coefficient determines the sensitivity of the dimension to distance changes. The bandwidth coefficients of the eight dimensions increase arithmetically to 4.0 starting from 0.5. Dividing the radial basis function value by 1 and adding the sum of the radial basis function values ​​yields the S-shaped probability mapping value. This mapping compresses the distance metric to the range of 0 to 0.5, so that extremely small distances produce outputs close to 0, and extremely large distances produce outputs close to 0.5, avoiding extreme values ​​from dominating the threshold calculation results. The contribution scaling factor is used to adjust the relative importance of each dimension in the threshold calculation. Based on the reconstruction contribution of each dimension during the training process of the variational autoencoder, the eight dimensions are assigned values ​​of 0.18, 0.22, 0.10, 0.15, 0.08, 0.12, 0.07, and 0.08, respectively, with a total of 1.0. The multiplicative exponential weight applies an exponential amplification or reduction to the sigmoid probability mapping value. Its value is set according to the significance of each dimension; significant dimensions are assigned values ​​greater than 1 to amplify the contribution, while insignificant dimensions are assigned values ​​less than 1 to suppress the contribution. The eight dimensions are assigned values ​​of 1.5, 1.2, 0.8, 1.1, 0.6, 0.9, 0.5, and 0.7, respectively. The multiplication operation multiplies the weighted probability mapping values ​​of all dimensions to obtain the comprehensive threshold value. Large deviations in any dimension will significantly raise the threshold through the multiplicative effect, achieving multiplicative fusion judgment of multidimensional anomalous evidence and avoiding threshold fluctuations caused by single-dimensional noise.

[0032] Preferably, in the sliding window integral walk test, the sliding window length is 60 sampling points, the window overlap rate is 50%, and the walk step size of the integral walk test is 5 sampling points; in the adversarial game constraint, the environmental stress weight is 0.55 to 0.65, the physical sign distortion weight is 0.35 to 0.45, and the sum of the two weights is always 1; the collaborative early warning level signal is divided into four levels, and the level division is based on the comparison between the weighted product of the environmental risk intensity component and the physical sign instability intensity component and the decision threshold; the decision threshold includes a high threshold value and a low threshold value, the high threshold value is 1.5 times the dynamic anomaly decision threshold, and the low threshold value is 0.6 times the dynamic anomaly decision threshold.

[0033] Preferred, such as Figure 2 As shown, the specific steps of the joint probability density reconstruction performed by the high-altitude hypoxia physical sign distortion identification model in S2 are as follows: S21, the multimodal physiological parameters of blood oxygen saturation, heart rate and respiratory rate are respectively truncated in the time domain, and the truncated length is aligned with the output timestamp of the perturbation tensor space; S22, based on the truncated parameter segments, the empirical cumulative distribution function value of each parameter in the current segment is calculated; S23, the empirical cumulative distribution function value of each parameter is projected to the high-dimensional joint probability density space, and a joint distribution surface is generated by using a kernel density estimation strategy; S24, the probability density offset relative to the reference distribution surface in the joint distribution surface is extracted, and the offset is accumulated to obtain the physical sign distortion indicator factor.

[0034] Specifically, in step S2, the joint probability density reconstruction of the high-altitude hypoxia sign distortion identification model involves the following steps: S21 first performs time-domain segment truncation on three physiological parameters: blood oxygen saturation, heart rate, and respiratory rate. The starting time of the truncation operation is based on the timestamp output of the perturbation tensor space, tracing back 120 sampling points as the segment start point and extending forward to the current time as the segment end point. Each segment includes 120 sampling data points, with a sampling frequency of 25 Hz corresponding to a time span of 4.8 seconds, which matches the human respiratory and circulatory coupling cycle. S22 calculates the empirical cumulative distribution function value for each parameter segment after truncation. During the calculation, the 120 data points in each segment are arranged in ascending order of value, and the sorted sequence is assigned an equally spaced cumulative probability increment. The cumulative probability value of each data point is its index divided by 120. For adjacent repeated values, the cumulative probability value corresponding to the largest index is taken as the cumulative distribution function value of that value. S23 constructs a three-dimensional coordinate point from the empirical cumulative distribution function values ​​of the three types of parameters at the same time. This coordinate point is projected onto a three-dimensional joint probability density space. A Gaussian kernel density estimation strategy is used to generate a joint distribution surface. All off-diagonal elements of the kernel function bandwidth matrix are set to zero, while the diagonal elements are set to 1.5% for blood oxygen saturation, 3 beats per minute for heart rate, and 1.5 breaths per minute for respiratory rate. The kernel function contributes density values ​​by spreading outward from each coordinate point in the three-dimensional space. S24 extracts the probability density value of the current three-dimensional coordinate point on the joint distribution surface. Simultaneously, it extracts the baseline probability density value at the same coordinate position on the pre-stored baseline distribution surface. The absolute offset between the current density value and the baseline density value is calculated. This offset is accumulated along the time axis moment by moment, with the accumulation window width set to 50 sampling points. The accumulated result is used as the vital sign distortion indicator factor output at that moment.

[0035] Preferred, such as Figure 3As shown, the specific steps of the variational autoencoder anomaly threshold coupler performing latent space probability distribution alignment in S3 are as follows: S31, the physical characteristic distortion indicator factor is encoded as a pair of mean and variance parameters in the latent space, and the encoding network of the variational autoencoder outputs the parameter pair; S32, latent variable samples are obtained by resampling from the Gaussian distribution described by the mean and variance parameter pair; S33, the KL divergence value between the latent variable sample and the pre-stored normal latent space distribution is calculated; S34, the KL divergence value is input into the threshold mapping function, and the dynamic anomaly decision threshold is output after the nonlinear transformation of the threshold mapping function.

[0036] Specifically, in step S3, the variational autoencoder anomaly threshold coupler performs latent space probability distribution alignment. In step S31, the vital sign distortion indicator factor is input into the encoding network of the variational autoencoder. The encoding network includes two parallel fully connected branches. The first branch outputs an 8-dimensional mean parameter vector, and the second branch outputs an 8-dimensional variance parameter vector. The weight matrix dimension of the fully connected layer is 64 times the number of input nodes multiplied by 8 times the number of output nodes, and the bias vector dimension is 8. In step S32, a resampling operation is performed from the multidimensional Gaussian distribution described by the mean and variance parameter vectors. During resampling, an 8-dimensional standard normal distribution random number vector is first generated. The square root of the variance parameter vector is then multiplied element-wise by the random number vector. The product is then added element-wise by the mean parameter vector to obtain the latent variable sample vector. This operation ensures that the gradient information remains continuous and differentiable during backpropagation. S33 calculates the KL divergence between the latent variable sample vector and the pre-stored normal latent space distribution. The normal distribution parameters include the baseline mean vector and the baseline covariance matrix. Each element of the baseline mean vector is obtained by averaging 600 samples of the encoded historical normal data. The baseline covariance matrix is ​​obtained by averaging the deviation matrices of these 600 samples. The KL divergence value is calculated by dividing the logarithm of the probability density of the latent variable sample vector by the expected value of the logarithm of the probability density of the normal distribution. S34 inputs the KL divergence value into the threshold mapping function, which is a composite function. First, the KL divergence value is multiplied by a gain factor of 0.25. Then, an exponential operation with the natural constant as the base is applied to this product. The result of the exponential operation is incremented by 1 and the reciprocal is taken. Then, 1 is subtracted from the reciprocal to obtain the mapping value. The mapping value is then multiplied by a scaling factor of 1.2 to obtain the dynamic anomaly decision threshold output.

[0037] Preferred, such as Figure 4As shown, the specific steps of the non-stationary temporal mutation detection and analysis platform in S4 for performing a sliding window integral walk test on the threshold sequence are as follows: S41, set a sliding window on the threshold sequence, and calculate the local integral area of ​​the threshold sequence in each window; S42, recursively push the local integral area forward according to the walk step size to generate the integral area walk trajectory; S43, perform a signed-rank test on the walk trajectory to obtain the deviation rank sum value of the integral area in each window relative to the median of the global integral area; S44, mark the window with the deviation rank sum value exceeding the preset walk threshold as a covariant mutation point window, and extract the spatiotemporal coordinates of the mutation point.

[0038] Specifically, in step S4, the non-stationary temporal abrupt change detection and analysis platform performs a sliding window integral walk test on the threshold sequence. In step S41, a sliding window is set on the threshold sequence, with a fixed length of 60 sampling points, corresponding to a 6-second data span. The initial window position is set at the first sampling point of the sequence. The threshold values ​​within each window are arranged in time index order. The absolute value of the deviation between the threshold values ​​of all sampling points within the window and the global mean of the entire sequence is calculated. This absolute value is multiplied by the sampling interval of 0.1 seconds and accumulated; the accumulated value is the local integral area of ​​the window. In step S42, the window is moved forward by 5 sampling points as the walk step. The integral area calculation is repeated at the new window position. The calculated integral areas are arranged according to the window sliding order, forming an integral area walk trajectory sequence. The walk process continues until the right boundary of the window reaches the end of the sequence. S43 performs a Wilcoxon signed-rank test on the wandering trajectory sequence. During the test, the median of the integral area of ​​the entire sequence is used as the central reference value. The difference between each integral area and the median is calculated for each item. The absolute values ​​of the differences are sorted in ascending order and assigned ranks. The sign of the original difference is added to the rank to obtain the signed-rank sum. S44 marks windows whose absolute values ​​of the signed-rank sum are greater than the critical value of 5.0 as deviation windows. The critical value of 5.0 corresponds to the critical value of the two-sided test at a significance level of 0.05. For each deviation window, the sampling point with the minimum time index is taken as the time coordinate of the mutation point. At the same time, the spatial coordinates are read from the perturbation tensor space grid position corresponding to the window. The two are combined to form the spatiotemporal coordinate output set of the covariant mutation point.

[0039] Preferred, such as Figure 5As shown, the specific steps of the collaborative early warning adjudication logic embedded in S5, which incorporates environmental stress weights and physical characteristic distortion weights in an adversarial game constraint, are as follows: S51, extract the current environmental stress intensity value and the current physical characteristic distortion intensity value based on the spatiotemporal coordinates of the mutation point; S52, multiply the environmental stress intensity value and the physical characteristic distortion intensity value by the environmental stress weight and the physical characteristic distortion weight, respectively, to obtain the weighted stress value and the weighted distortion value; S53, calculate the difference between the weighted stress value and the weighted distortion value, and use this difference as the game adversarial deviation; S54, based on the numerical range of the game adversarial deviation, adjudicate and output the corresponding early warning level signal.

[0040] Specifically, in step S5, the collaborative early warning adjudication logic is embedded in the specific sub-steps of the adversarial game constraint between environmental stress weights and physical sign distortion weights. In step S51, based on the timestamp information in the spatiotemporal coordinates of the mutation point, the magnitude of the environmental stress gradient vector corresponding to that timestamp is retrieved from the memory cache of the environmental disturbance feature mapping module, and the value of the physical sign distortion indicator factor corresponding to the same timestamp is retrieved from the memory cache of the physical sign distortion identification module. If the timestamp exists in both modules' caches, it is read directly; if only adjacent timestamp data exists, linear interpolation is used to estimate the value of the target timestamp. In step S52, the environmental stress intensity value is multiplied by the environmental stress weight factor 0.60 to obtain the weighted stress value, and the physical sign distortion intensity value is multiplied by the physical sign distortion weight factor 0.40 to obtain the weighted distortion value. The sum of the two weight factors is always equal to 1.0, and a normalization check is performed after each adjustment to ensure that the values ​​of both factors are within the closed interval of 0 to 1. S53 calculates the difference between the weighted stress value and the weighted distortion value. This difference is defined as the game adversarial bias. A positive value indicates that environmental stress dominates the early warning trigger, while a negative value indicates that physical distortion dominates the early warning trigger. The absolute value of the bias represents the degree of dominance. S54 determines the early warning level based on the numerical range of the bias. If the bias is greater than 0.15, a level three signal characterized by environmental dominance is output. If the bias is less than -0.15, a level three signal characterized by physical dominance is output. If the bias is between -0.15 and 0.15 and the sum of the weighted stress value and the weighted distortion value is greater than 0.8 times the dynamic anomaly judgment threshold, a level two signal is output. If the sum is less than or equal to 0.8 times the dynamic anomaly judgment threshold, a level one signal is output. After the determination, the sign of the current bias is used as the basis for the next weight factor adjustment direction.

[0041] A collaborative monitoring and early warning method for personnel vital signs and the environment in high-risk energy scenarios on plateaus is proposed. This method is implemented through a collaborative monitoring and early warning system for personnel vital signs and the environment in high-risk energy scenarios on plateaus, comprising: a plateau environmental disturbance feature mapping module, a high-altitude hypoxia vital sign distortion identification module, a variational autoencoder anomaly threshold coupling module, a non-stationary temporal mutation detection and analysis module, a collaborative early warning adjudication game module, and an early warning level output module. The plateau environmental disturbance feature mapping module receives externally acquired temporal data of air pressure, oxygen partial pressure, and temperature and humidity, and outputs a disturbance tensor space to the high-altitude hypoxia vital sign distortion identification module. The high-altitude hypoxia vital sign distortion identification module receives the output of the disturbance tensor space and integrates externally acquired multimodal physiological parameters such as blood oxygen saturation, heart rate, and respiratory rate. The system outputs a vital sign distortion indicator factor to the variational autoencoder anomaly threshold coupling module; the variational autoencoder anomaly threshold coupling module performs latent space probability distribution alignment on the vital sign distortion indicator factor and outputs a dynamic anomaly judgment threshold to the non-stationary temporal mutation detection and analysis module; the non-stationary temporal mutation detection and analysis module performs a sliding window integral walk test on the threshold sequence and outputs the spatiotemporal coordinates of the covariant mutation point to the collaborative early warning adjudication game module; the collaborative early warning adjudication game module embeds an adversarial game constraint of environmental stress weight and vital sign distortion weight based on the spatiotemporal coordinates of the mutation point and outputs an early warning level signal to the early warning level output module; the early warning level output module decomposes the early warning level signal into an environmental hazard intensity component and a vital sign instability intensity component and presents them externally.

[0042] This invention proposes a collaborative monitoring and early warning method for personnel vital signs and the environment in high-risk energy scenarios on plateaus. It addresses the traditional model of separate environmental and vital sign monitoring by cascading and coupling an environmental disturbance feature mapping algorithm and a high-altitude, low-oxygen vital sign distortion identification model. This accurately projects the environmental stress gradient to the vital sign distortion identification stage, giving vital sign analysis an environmentally relevant perspective. It overcomes the shortcomings of traditional methods that treat the environment as an exogenous variable and ignore its dynamic driving effect on the transmission of vital sign distortion. The variational autoencoder anomaly threshold coupler aligns the probability distribution of vital sign distortion indicators in the latent space, dynamically generating a decision threshold that adaptively adjusts to the degree of environmental disturbance and the drift of the vital sign baseline. This completely replaces the fixed threshold setting method in existing technologies, effectively suppressing frequent false alarms during periods of severe environmental fluctuations.

[0043] The mutation detection and game-theoretic adjudication stages of this method further enhance the reliability and flexibility of the early warning system. The non-stationary temporal mutation detection and analysis platform, by performing a sliding window integral walk test on the dynamic threshold sequence, can accurately locate the covariant mutation point between environmental disturbances and vital sign drift, avoiding serious missed warnings caused by the accumulation of slowly changing stresses, and compensating for the insufficient adaptability of existing static threshold strategies in non-stationary scenarios. The collaborative early warning adjudication logic introduces adversarial game constraints between environmental stress weights and vital sign distortion weights, so that the early warning level judgment is no longer a simple weighted superposition of two types of parameters, but rather a dynamic game-theoretic balance of their dominant roles in triggering the early warning, taking into account the dual effects of environmental hazard intensity and vital sign instability severity, outputting an early warning level signal including dual-component information, providing a reliable technical means for full-process adaptive collaborative monitoring in high-risk scenarios of plateau energy.

[0044] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," "link," and "fix" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0045] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for coordinated monitoring and early warning of personnel vital signs and the environment in high-risk energy scenarios on plateaus, characterized in that, Includes the following steps: S1, construct a plateau environment disturbance feature mapping algorithm, project the collected air pressure, oxygen partial pressure and temperature and humidity time series data into the disturbance tensor space, and extract the environmental stress gradient; S2, based on the perturbation tensor space output, starts the high-altitude hypoxia sign distortion identification model, performs joint probability density reconstruction on multimodal physiological parameters such as blood oxygen saturation, heart rate and respiratory rate, and generates sign distortion indicator factors. S3, input the distortion indicator factor into the variational autoencoder anomaly threshold coupler, and calculate the dynamic anomaly decision threshold through the latent space probability distribution alignment operation; S4 utilizes a non-stationary time-series mutation detection and analysis platform to perform a sliding window integral walk test on the threshold sequence, capturing covariant mutation points between environmental disturbances and vital sign drift; S5, based on the spatiotemporal coordinates of the mutation point, activates the collaborative early warning adjudication logic, and embeds the adjudication logic with the adversarial game constraint of environmental stress weight and physical sign distortion weight; S6 outputs a collaborative monitoring and early warning level signal, which includes the environmental risk intensity component and the physical instability intensity component.

2. The method for coordinated monitoring and early warning of personnel vital signs and environment in high-risk energy scenarios on plateaus as described in claim 1, characterized in that, The expression for the plateau environment disturbance feature mapping algorithm is as follows: ;in, It represents the cumulative potential of environmental disturbances. For the spatial integration field, It is a spatial position vector. The pressure field distribution function is... The enthalpy-humidity coupling potential function is... For time variables, This is the pressure gradient weighting coefficient. The enthalpy-humidity time-varying coupling coefficient is... For perturbation mapping quantization value, The total number of sampling points. For sampling point index, It is a nonlinear modulation factor. It is the stability constant. For the first The spatial weighted index of each sampling point, For the first The cumulative potential of the perturbation at each sampling point is taken as follows: For gradient operators, This is the symbol for partial differentials.

3. The method for coordinated monitoring and early warning of personnel vital signs and environment in high-risk energy scenarios on plateaus as described in claim 1, characterized in that, The expression for the high-altitude hypoxia physical sign distortion recognition model is: ;in, For multimodal physiological parameter joint bias functional, This represents the total number of physiological parameter categories. For category indexing, For the first Deviation weights of class parameters For the first Real-time observations of class parameters For the first The baseline center value of the class parameter, For the first The baseline scaling factor for class parameters For the first The saturation suppression coefficient of the class parameter, For the first The sliding difference value of the class parameter, For the first Differential scaling constants for class parameters, The cumulative response for distortion identification. This is the upper limit of the integration time window. For integration time variable, for The joint bias functional values ​​at time t, It is the saturation cutoff constant. for The decay function of ambient oxygen partial pressure at time t. This is the differential symbol.

4. The method for coordinated monitoring and early warning of personnel vital signs and environment in high-risk energy scenarios on plateaus as described in claim 1, characterized in that, The expression for the variational autoencoder anomaly threshold coupler is: ,in, This is the threshold value for dynamic anomaly detection. The total number of dimensions encoded in the latent space. For dimensional indexing, for Dimensional contribution scaling factor For the variational autoencoder to encode the distortion indicator factor, at the 1st... The values ​​of the latent variables in the dimension, For the first The baseline cluster center vector of dimension, For the Euclidean norm, For the first The radial bandwidth coefficient of the dimension, For the first Multiplicative exponent weights of dimensions.

5. The method for coordinated monitoring and early warning of personnel vital signs and environment in high-risk energy scenarios on plateaus as described in claim 1, characterized in that, In the sliding window integral walk test, the sliding window length is 60 sampling points, the window overlap rate is 50%, and the walk step size is 5 sampling points. In the adversarial game constraint, the environmental stress weight is 0.55 to 0.65, the physical characteristic distortion weight is 0.35 to 0.45, and the sum of the two weights is always 1. The collaborative early warning level signal is divided into four levels, and the level division is based on the comparison between the weighted product of the environmental risk intensity component and the physical characteristic instability intensity component and the decision threshold. The decision threshold includes a high threshold value and a low threshold value. The high threshold value is 1.5 times the dynamic anomaly decision threshold value, and the low threshold value is 0.6 times the dynamic anomaly decision threshold value.

6. The method for coordinated monitoring and early warning of personnel vital signs and environment in high-risk energy scenarios on plateaus as described in claim 1, characterized in that, The joint probability density reconstruction performed by the S2 high-altitude hypoxia physical sign distortion identification model includes: The multimodal physiological parameters of blood oxygen saturation, heart rate and respiratory rate are truncated in the time domain, and the truncated length is aligned with the output timestamp of the perturbation tensor space. Based on the extracted parameter segments, calculate the empirical cumulative distribution function value of each parameter within the current segment; The empirical cumulative distribution function values ​​of each parameter are projected onto a high-dimensional joint probability density space, and a joint distribution surface is generated using a kernel density estimation strategy. Extract the probability density offset relative to the reference distribution surface from the joint distribution surface, and sum the offsets to obtain the characteristic distortion indicator factor.

7. The method for coordinated monitoring and early warning of personnel vital signs and environment in high-risk energy scenarios on plateaus as described in claim 1, characterized in that, The latent space probability distribution alignment performed by the variational autoencoder anomaly threshold coupler in S3 includes: The indicator factors of physical deformity are encoded as a pair of mean and variance parameters in the latent space, and the encoding network of the variational autoencoder outputs the parameter pair; latent variable samples are obtained by resampling from the Gaussian distribution described by the mean and variance parameter pair. Calculate the KL divergence value between the latent variable sample and the pre-stored normal latent space distribution; input the KL divergence value into the threshold mapping function, and output the dynamic anomaly decision threshold after nonlinear transformation of the threshold mapping function.

8. The method for coordinated monitoring and early warning of personnel vital signs and environment in high-risk energy scenarios on plateaus as described in claim 1, characterized in that, The non-stationary temporal abrupt change detection and analysis platform in S4 performs a sliding window integral walk test on the threshold sequence, including: Set a sliding window on the threshold sequence, and calculate the local integral area of ​​the threshold sequence within each window; The local integral area is recursively pushed forward according to the walking step size to generate the integral area walking trajectory; the signed rank test is performed on the walking trajectory to obtain the deviation rank sum value of the integral area in each window from the median of the global integral area; the window with the deviation rank sum value exceeding the preset walking threshold is marked as the covariant mutation point window, and the spatiotemporal coordinates of the mutation point are extracted.

9. The method for coordinated monitoring and early warning of personnel vital signs and environment in high-risk energy scenarios on plateaus as described in claim 1, characterized in that, The collaborative early warning and adjudication logic in S5 incorporates environmental stress weights and vital sign distortion weights in an adversarial game constraint, including: Extract the current environmental stress intensity value and the current physical characteristic distortion intensity value based on the spatiotemporal coordinates of the mutation point; multiply the environmental stress intensity value and the physical characteristic distortion intensity value by the environmental stress weight and the physical characteristic distortion weight, respectively, to obtain the weighted stress value and the weighted distortion value; calculate the difference between the weighted stress value and the weighted distortion value, and use this difference as the game adversarial bias; Based on the numerical range of the game adversarial deviation, the adjudicator outputs the corresponding warning level signal.

10. The method for coordinated monitoring and early warning of personnel vital signs and environment in high-risk energy scenarios on plateaus, as described in any one of claims 1-9, is characterized in that... This method is implemented through a collaborative monitoring and early warning system for personnel vital signs and the environment in high-risk energy scenarios on plateaus, including: The system comprises the following modules: Plateau Environmental Disturbance Feature Mapping Module, Altitude Cold and Hypoxia Symptom Distortion Recognition Module, Variational Autoencoder Abnormal Threshold Coupling Module, Non-Stationary Temporal Abrupt Change Detection and Analysis Module, Collaborative Early Warning Decision Game Theory Module, and Early Warning Level Output Module. The Plateau Environmental Disturbance Feature Mapping Module receives externally collected temporal data on air pressure, oxygen partial pressure, and temperature and humidity, and outputs a disturbance tensor space to the Altitude Cold and Hypoxia Symptom Distortion Recognition Module. The Altitude Cold and Hypoxia Symptom Distortion Recognition Module receives the output of the disturbance tensor space and connects to externally collected multimodal physiological parameters such as blood oxygen saturation, heart rate, and respiratory rate, and outputs symptom distortion indicator factors to the Variational Autoencoder Abnormal Threshold Coupling Module. The variational autoencoder anomaly threshold coupling module performs latent space probability distribution alignment on the vital sign distortion indicator factors and outputs the dynamic anomaly judgment threshold to the non-stationary temporal mutation detection and analysis module. The non-stationary temporal mutation detection and analysis module performs a sliding window integral walk test on the threshold sequence and outputs the spatiotemporal coordinates of the covariant mutation point to the collaborative early warning adjudication game module. The collaborative early warning adjudication game module embeds the adversarial game constraint of environmental stress weight and vital sign distortion weight based on the spatiotemporal coordinates of the mutation point and outputs the early warning level signal to the early warning level output module. The early warning level output module decomposes the early warning level signal into environmental hazard intensity component and vital sign instability intensity component and presents them externally.

Images