A big data-based user health monitoring management method and system

Abstract

The application discloses a user health monitoring management method and system based on big data, relates to the technical field of big data health monitoring, and comprises the following steps: collecting user multi-modal physiological data, preprocessing and resampling the multi-modal physiological data to form a health analysis sequence, analyzing original candidate embedding dimensions according to the health analysis sequence to construct a phase space state vector, calculating space indexes based on the phase space state vector, including local trajectory divergence intensity, phase space neighborhood distribution entropy and trajectory energy activity, establishing a robust baseline according to the space indexes and mapping to obtain a single-mode chaotic vitality value, forming a chaotic vitality sequence from the single-mode chaotic vitality value and taking the chaotic vitality sequence as a time-delay causal network node sequence, establishing a node self-history state based on the node sequence, and selecting a node pair to establish a source node lag state and a target node next window state. The application significantly improves the accuracy, time sequence sensitivity and interpretability of health monitoring.

Description

Technical Field

[0001] This invention relates to the field of big data health monitoring technology, and in particular to a user health monitoring and management method and system based on big data. Background Technology

[0002] With the rapid development of wearable devices, IoT sensors, and mobile health terminals, big data-based user health monitoring and management technologies have gradually become an important research direction in the fields of digital healthcare and smart health management. Existing technologies typically involve continuous collection of multi-source physiological signals (such as heart rate, blood oxygen saturation, respiratory rate, skin conductance, and body movement information), combined with preprocessing steps such as data cleaning, denoising, resampling, and feature extraction to construct health status assessment models. In terms of modeling methods, traditional approaches often employ statistical indicators (mean, variance, coefficient of variation, etc.), machine learning classification models (such as support vector machines and random forests), or deep learning networks (such as LSTM and Transformer) to identify health status. However, existing technologies still have shortcomings in practical applications. Most methods remain within the framework of "single-modal independent analysis or simple fusion," failing to fully characterize the dynamic coupling relationships and temporal dependencies between different physiological systems, resulting in insufficient ability to identify overall imbalances in complex physiological systems. Summary of the Invention

[0003] In view of the aforementioned existing problems, the present invention is proposed.

[0004] Therefore, this invention provides a user health monitoring and management method and system based on big data, which solves the problem that existing technologies are insufficient in recognizing the overall synergistic imbalance of complex physiological systems.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a user health monitoring and management method based on big data, comprising, Multimodal physiological data of users are collected, preprocessed and resampled to form a health analysis sequence, and the original candidate embedding dimension is analyzed based on the health analysis sequence to construct a phase space state vector; Spatial indices, including local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity, are calculated based on phase space state vectors. A robust baseline is established based on the spatial indices and mapped to obtain the single-mode chaotic activity value. The single-modal chaotic vitality values ​​are formed into a chaotic vitality sequence and used as a time-delayed causal network node sequence. Based on the node sequence, the node self-historical state is established. Node pairs are selected to establish the source node's lag state and the target node's next window state. Triples are established by combining the node self-historical state and the node pairs are analyzed to generate time-delayed causal edges. A causal adjacency matrix is ​​established based on time-delayed causal edges, and a basic health score and a short-term deterioration trend penalty term are calculated. A causal chaotic health index is calculated by combining the basic health score and the short-term deterioration trend penalty term. The user's health status is identified based on the causal chaotic health index, and an intervention trigger marker is generated.

[0006] As a preferred embodiment of the user health monitoring and management method based on big data described in this invention, the step of collecting user multimodal physiological data for preprocessing and resampling to form a health analysis sequence specifically involves: collecting raw user multimodal physiological data through intelligent monitoring equipment. It includes four types of physiological modal data: heart rate variability modality (h), skin conductance modality (s), respiratory modality (r), and electroencephalogram (EEG) modality (b). Define the length of the health analysis window and update step size Define the k-th health analysis window ; The raw multimodal physiological data falling into the health analysis window are extracted and preprocessed for each modality. The preprocessed multimodal physiological data were uniformly resampled to the health standard time axis to obtain resampled standard modality data. And form a standardized multimodal health analysis sequence. .

[0007] As a preferred embodiment of the user health monitoring and management method based on big data described in this invention, the step of constructing the phase space state vector based on the original candidate embedding dimension analysis of the health analysis sequence specifically involves: for the j-th type of standard modality data sequence Slow-varying trend terms are obtained by using fixed-length median trend filtering; From standard modal data sequences Subtracting the slow-changing trend term from the middle yields the health dynamic sequence. ; The health dynamic sequence is divided into equal-frequency binning methods. A range of amplitude values ​​is defined, and candidate delays are defined based on these ranges. Calculate the average mutual information of candidate delays; The first local minimum is selected as the candidate delay time based on the average mutual information. ; Define candidate embedding dimensions And construct a temporary phase space vector based on the candidate embedding dimension; For each temporary phase space vector, find the nearest neighbor index. The proportion of false nearest neighbors is calculated based on the nearest neighbor index and the temporary phase space vector. Determine the original candidate embedding dimension based on the proportion of false nearest neighbors; Construct the phase space state vector using the original candidate embedding dimension. The phase space state vectors of all phase spaces are combined to form the set of phase space trajectories of the j-th mode data. .

[0008] As a preferred embodiment of the user health monitoring and management method based on big data described in this invention, the calculation of spatial indicators based on phase space state vectors specifically involves determining the number of constructable phase space state vectors according to the original candidate embedding dimensions. Determine the effective index set for the future l-step size of the phase space based on the number of constructible phase space state vectors. ; Calculate the average divergence distance of the phase space trajectory of the nearest neighbor point after several phase space steps based on the nearest neighbor index and the effective index set; The local trajectory divergence intensity is obtained by linearly fitting the average divergence distance; Constructing the collection of user historical phase space trajectories For each phase space state vector in the historical phase space trajectory set, find the nearest neighbor distance; The individual baseline neighborhood radius is obtained by calculating the median distance of the nearest neighbor distances of all phase space state vectors in the historical phase space trajectory set and multiplying it by a neighborhood magnification factor. ; For each phase space state vector in the current window phase space trajectory, the number of neighborhood occupancy is counted and normalized to the neighborhood distribution probability. The phase space neighborhood distribution entropy is calculated based on the neighborhood distribution probability. The trajectory velocity between adjacent phase space state vectors is calculated based on the current window's phase space trajectory set, and the trajectory energy activity is calculated based on the trajectory velocity.

[0009] As a preferred embodiment of the user health monitoring and management method based on big data described in this invention, the step of establishing a robust baseline based on spatial indicators and mapping it to obtain a single-modal chaotic vitality value specifically involves: establishing a robust baseline in the historical health window using the median and median absolute deviation based on any one of the indicators among local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity. Based on the robust baseline, robust deviation values ​​are obtained by robustly normalizing the local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity. The vitality component is obtained by mapping the robust deviation value using an S-shaped mapping function; The single-mode chaotic vitality value is obtained by weighting all vitality components. .

[0010] As a preferred embodiment of the user health monitoring and management method based on big data described in this invention, the following steps are taken: The single-modal chaotic vitality values ​​are used to form a chaotic vitality sequence and serve as a time-delay causal network node sequence. The establishment of the node's historical state based on the node sequence specifically involves: forming a network construction window with the current window k as the endpoint. In the network construction window, the single-mode chaotic vitality values ​​of each mode are formed into a chaotic vitality sequence, and the chaotic vitality sequence is used as the node sequence of the time-delay causal network. The node sequence is converted into a three-state health regulation symbol sequence, and a second-order self-history state is constructed for each node based on the three-state health regulation symbol sequence.

[0011] As a preferred embodiment of the user health monitoring and management method based on big data described in this invention, the step of selecting node pairs to establish the source node's lagging state and the target node's next window state, and combining the node's historical state to establish triples and analyze the generation of time-delayed causal edges of node pairs specifically involves: selecting any two different nodes to define the source node. and target node Establish the source node's lag state and the target node's next window state respectively; Based on the source node's lag state, the target node's next window state, and the target node's second-order self-historical state, a window is constructed to count the ternary states, and the number of occurrences of each ternary state is counted in the network construction window. Calculate the conditional probability of the target node's next state under the condition that the source node participates and the conditional probability of the target node's next state without considering the source node based on the number of occurrences of the ternary state. Further, calculate the joint probability of the number of occurrences of the ternary state in the current network construction window. Calculate the source node based on conditional probability and joint probability. In time lag Next target node The conditional transition entropy; Calculate the source node based on the node sequence. To the target node In time lag Consistency of the lagged correlation direction; Candidate causal edge symbols are determined based on directional consistency, including , And the three symbols: 0; The evidence value of candidate causal edges is calculated by combining conditional transition entropy and directional consistency. For each pair of nodes arrive The time lag with the largest candidate causal edge evidence value is selected as... arrive Optimal causal time delay The system extracts the corresponding maximum candidate causal edge evidence value, calculates the historical maximum candidate causal evidence value set, and extracts the median and median absolute deviation from the historical maximum candidate causal evidence value set for weighted summation to obtain the individualized edge evidence threshold. If the maximum candidate causal edge evidence value is greater than the individualized edge evidence threshold and the candidate causal edge sign is not 0, then a node is generated. arrive Determine the time-delayed causal edges and establish a set of causal edges. ; Calculate the final edge weight of the time-delayed causal edge.

[0012] As a preferred embodiment of the user health monitoring and management method based on big data described in this invention, the step of establishing a causal adjacency matrix based on time-delayed causal edges and calculating a basic health score and a short-term deterioration trend penalty term, and calculating a causal chaotic health index by combining the basic health score and the short-term deterioration trend penalty term, specifically involves: calculating the causal edge strength based on each causal edge in the causal edge set and constructing a time-delayed causal adjacency matrix; The causal input flow and causal output flow of a node are calculated based on the time-delay causal adjacency matrix, and the net causal flow of the node is obtained by subtracting the causal output flow from the causal input flow. Calculate the causal density and average causal propagation strength of the network based on the causal edge set; The causal network load is obtained by weighting the network causal density and the average causal propagation strength. Calculate the causal source concentration based on the net causal flow; Weighting the causal source concentration and causal network load and truncating the result in a causal network anomaly risk within the range of 0 to 1, and subtracting the causal network anomaly risk from 1, yields the causal network structural health component. ; The total chaos vitality value is obtained by weighting all single-mode chaos vitality values. The total chaos vitality value is then normalized and truncated to obtain the risk of insufficient chaos vitality. The chaotic vitality health component is obtained by subtracting the risk of insufficient chaotic vitality from 1. The health component of the causal network structure And Chaos Vitality Health Quantity The basic health score is obtained by weighted summation. ; Define the number of trend retracing periods The decline rate of the chaotic vitality health component and the decline rate of the causal network structure health component were calculated based on the number of trend review cycles. The short-term deterioration trend penalty term is obtained by weighted summing of the decline in the chaotic vitality health component and the decline in the causal network structure health component and truncating it in the range of 0 to 1. The causal chaotic health index is calculated based on the short-term deterioration trend penalty term.

[0013] As a preferred embodiment of the user health monitoring and management method based on big data described in this invention, the step of identifying user health status and generating intervention trigger markers based on causal chaotic health index specifically involves identifying user health status based on causal chaotic health index, including green stable state, yellow compensated state, and red decompensated state. An intervention trigger marker is generated for each user's health status, including 0, 1, and 2, which correspond to the green stable status, the yellow compensated status, and the red decompensated status, respectively. Corresponding intervention measures are taken for each intervention trigger marker.

[0014] Secondly, this invention provides a user health monitoring and management system based on big data, comprising, The raw data analysis module is used to collect user multimodal physiological data, preprocess and resample it to form a health analysis sequence, and construct a phase space state vector based on the analysis of the original candidate embedding dimension of the health analysis sequence. The Chaotic Vitality Mapping Module is used to calculate spatial indices based on phase space state vectors, including local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity. It establishes a robust baseline based on the spatial indices and maps them to obtain the single-mode chaotic vitality value. The causal network construction module is used to form a chaotic vitality sequence from the single-modal chaotic vitality value and use it as a time-delayed causal network node sequence. Based on the node sequence, the self-historical state of the node is established. The source node's lagging state and the target node's next window state are selected by node pairs. Triples are established by combining the node's self-historical state and the node pairs are analyzed to generate time-delayed causal edges. The health analysis module is used to establish a causal adjacency matrix based on time-delayed causal edges and calculate a basic health score and a short-term deterioration trend penalty term. It then calculates a causal chaotic health index by combining the basic health score and the short-term deterioration trend penalty term, identifies the user's health status based on the causal chaotic health index, and generates an intervention trigger marker.

[0015] The beneficial effects of this invention are as follows: By collecting multimodal physiological data and performing preprocessing and resampling, a unified health analysis sequence is constructed, and a state vector is generated based on phase space reconstruction, thereby effectively expressing the nonlinear dynamic characteristics of complex physiological systems and improving the structured representation capability of the original time-series data. Furthermore, by calculating the local trajectory divergence intensity, neighborhood entropy, and trajectory energy activity and establishing a robust baseline, a single-modal chaotic vitality value is obtained, realizing the normalization of individual differences and improving cross-user comparability. Then, the chaotic vitality sequence is used as a time-delayed causal network node, and the lagged state and triplet modeling method is introduced to characterize the temporal causal propagation relationship between multimodalities. Finally, a causal chaotic health index is constructed by combining the causal adjacency matrix with the basic health score and the short-term deterioration penalty term, realizing the dynamic quantitative assessment and risk classification of health status and intervention triggering, thereby significantly improving the accuracy, temporal sensitivity, and interpretability of health monitoring. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart of the user health monitoring and management method based on big data in Example 1.

[0018] Figure 2 This is a structural diagram of the user health monitoring and management system based on big data in Example 1. Detailed Implementation

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0020] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0021] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0022] Example 1, referring to Figure 1 and Figure 2 This is the first embodiment of the present invention, which provides a user health monitoring and management method based on big data, including the following steps: S1. Collect user multimodal physiological data, preprocess and resample to form a health analysis sequence, and construct a phase space state vector based on the original candidate embedding dimension analysis of the health analysis sequence. S1.1 Collect raw multimodal physiological data of users through intelligent monitoring equipment. The data includes four physiological modalities: heart rate variability (h), skin conductance (s), respiratory mode (r), and electroencephalography (b). Specifically, heart rate variability is acquired jointly by a wrist photoplethysmography (PPG) sensor and a chest strap ECG sensor, with the R-peak interval of the chest strap ECG being the preferred data source. Skin conductance is acquired by collecting skin conductance signals from the fingers or wrist using a smart ring or wrist electrode. Respiratory mode is acquired by collecting respiratory waveforms using a thoracic impedance sensor in a smart chest strap. EEG is acquired by collecting frontal or occipital EEG signals using a dry electrode head-mounted EEG device and extracting the power envelope in the 8Hz to 13Hz frequency band. Heart rate variability can reflect the state of autonomic nervous system regulation, skin conductance can reflect the level of sympathetic nerve excitation, respiratory waveform can reflect the state of rhythm regulation, and electroencephalogram (EEG) alpha wave power can reflect alertness, relaxation, and fatigue-related states. These four factors respectively cover cardiovascular, autonomic nervous, respiratory regulation, and central nervous system activities. Define the length of the health analysis window and update step size Define the k-th health analysis window ,in This is the window end time; Specifically, the states of fatigue, stress, and recovery ability in user health management are not instantaneous changes. It usually takes several minutes for the physiological regulatory structure to be reflected. Therefore, the window is set at 300 seconds, and the step size is 60 seconds. This is because updating the health status once per minute can meet the needs of real-time monitoring and management without causing frequent calculations and false alarms. S1.2 Extract the raw multimodal physiological data falling into the window according to the health analysis window and preprocess the raw physiological data of each modality; Specifically, for heart rate variability modal data h, R peaks are detected from chest cuff ECG signals and RR interval sequences are formed. If a certain RR interval sequence exceeds the normal heart rate range, linear interpolation of adjacent effective RR intervals is used to complete it. The normal heart rate range is 300ms-2000ms, corresponding to the heart rate range of adults at rest and under mild activity. For the skin conductance mode s, a low-pass filter is used to preserve the slowly changing skin conductance level. The system cutoff frequency is set to 2Hz. Both skin conductance level and skin conductance response belong to low-frequency autonomic nervous activity. Components above 2Hz are mostly motion artifacts or contact noise. For respiratory mode r, bandpass filtering is used to retain waveform components within the normal respiratory frequency range. The preferred normal respiratory frequency range is 0.2Hz-0.7Hz, corresponding to approximately 12 to 42 breaths per minute, covering common respiratory rhythms under resting, mild activity, and stress conditions, while eliminating extremely low-frequency drift and high-frequency motion noise. For EEG modality b, the raw EEG modality data is bandpass filtered from 8Hz to 13Hz, and then the alpha wave power envelope is calculated. The alpha wave power envelope is chosen as the EEG input because the health management scenario focuses on the user's alertness, relaxation and fatigue-related states, rather than the raw EEG waveform itself. The power envelope is more stable and more suitable for long-term wearable monitoring. The preprocessed multimodal physiological data were uniformly resampled to the health standard time axis to obtain resampled standard modality data. And form a standardized multimodal health analysis sequence. The health standard time axis is based on a standard sampling frequency. Obtain, Health Standard Timeline Represented as:

[0023] Resampling is mapped using cubic spline interpolation or equivalent interpolation methods, with the number of uniform sampling points in the k-th window being... ; The reason for choosing 4Hz as the standard sampling frequency is that fatigue, stress and regulatory function changes in user health management are mainly manifested as low-frequency dynamic processes. 4Hz is sufficient to describe changes in heart rate variability, skin conductance, respiratory rhythm and EEG alpha wave envelope. At the same time, 4Hz can significantly reduce the burden of subsequent phase space reconstruction and data calculation.

[0024] S1.3, For the j-th type of standard modal data sequence The slow-varying trend term is obtained by using a fixed-length median trend filter:

[0025] in Let m be the slow-varying trend term of the j-th type of standard modal data at the m-th sampling point. For median operations, The trend filter half-window length is preferably 20 sampling points, corresponding to 5 seconds. This time scale can reduce sensor drift and slow changes in posture, while not erasing dynamics related to health status such as respiratory rhythm, heart rate regulation and EEG alpha wave envelope. From standard modal data sequences Subtracting the slow-varying trend term from the middle yields the healthy dynamic sequence used for phase space reconstruction. Among them, the health dynamic sequence needs to be subject to boundary constraints. The upper and lower boundaries can be obtained from the health samples of the user's health status. The 1st percentile and 99th percentile of the historical health samples are taken as the upper and lower boundaries and the boundary constraints are executed. The health dynamic sequence is divided into equal-frequency binning methods. A range of amplitude values ​​is defined, and candidate delays are defined based on these ranges. Calculate the average mutual information of candidate delays:

[0026] in The delay of the j-th modal data in the k-th window is... The average mutual information over time, where c and d are the amplitude intervals. For the m-th sampling point The probability of falling into the c-th amplitude interval, For the first sampling points The probability of falling into the d-th amplitude interval, Then it means Falling into the c-th amplitude interval and The joint probability of falling into the d-th amplitude interval. To prevent extremely small integers like zero, the value is set to... ; The calculation is done by dividing the number of sampling points falling into the c-th amplitude interval by the total number of sampling points. Similarly... The result is calculated by dividing the number of sampling points falling into the d-th amplitude interval by the total number of sampling points. Then based on the sampling points and The number of times the sample falls within the corresponding amplitude intervals c and d is divided by the total number of sampling points. The preferred number of amplitude intervals is 16, and the number of sampling points within the window is 1200. 16 intervals can ensure that each interval has a sufficient number of samples and can distinguish the amplitude state changes of physiological signals. If the number of bins is too small, the mutual information resolution will be insufficient; if the number of bins is too large, the probability estimation will be unstable. The first local minimum is selected as the candidate delay time based on the average mutual information. If no local minimum exists within the search range, the global minimum is used as the original candidate delay time. A local minimum refers to a delay between the lower and upper limits of the delay search. The average mutual information is less than and The average mutual information of the delay is preferably 2 sampling points for the lower limit of the delay search and 40 sampling points for the upper limit. S1.4, Define candidate embedding dimensions And construct a temporary phase space vector based on the candidate embedding dimension:

[0027] in Let be the temporary phase space vector of the j-th modal data in the k-th window with embedding dimension d. This represents the health dynamic data for the i-th sampling point; For each temporary phase space vector, find the nearest neighbor index. When searching for the nearest neighbor index, it is necessary to exclude nodes with a time distance less than [a certain value]. Points at adjacent time points are naturally similar. If this is not excluded, consecutive points on the same trajectory may be misjudged as true phase space neighbors, thus underestimating the embedding dimension. Calculate the proportion of false nearest neighbors based on the nearest neighbor index and the temporary phase space vector:

[0028] in Let represent the proportion of false nearest neighbors for the j-th modality data in the k-th window with candidate embedding dimension d. This represents the number of phase space points available for computation under the candidate embedding dimension d. This is an indicator function that takes the value 1 when the condition in parentheses is true, and 0 otherwise. The threshold for identifying false nearest neighbors is preferably set to 10. When the ratio is greater than 10, it means that the two points appear to be close in d-dimensional space, but the distance increases significantly after adding a delay dimension, which is a false nearest neighbor caused by low-dimensional folding. Determine the original candidate embedding dimension based on the proportion of spurious nearest neighbors:

[0029] in For the original candidate embedding dimension, The threshold for the proportion of false nearest neighbors is preferably set to 0.05, which means that when the proportion of false nearest neighbors is less than 5%, the phase space is considered to have been basically expanded. in The lower bound for the embedded dimension search is set to 3. To set the upper limit of the embedding dimension search, a value of 8 is set. It is difficult to expand the nonlinear state of the human physiological regulation system below 3 dimensions, and above 8 dimensions will significantly increase the amount of computation. At the same time, it is easily affected by noise and window sample size limitations in wearable health monitoring scenarios. S1.5 Construct the phase space state vector using the original candidate embedding dimensions. The phase space state vectors of all phase spaces are combined to form the set of phase space trajectories of the j-th mode data. .

[0030] S2. Calculate spatial indices based on phase space state vectors, including local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity. Establish a robust baseline based on the spatial indices and map it to obtain the single-mode chaotic activity value. S2.1 Determine the number of constructible phase space state vectors based on the original candidate embedding dimensions:

[0031] in The number of phase space state vectors that can be constructed within the k-th window for the j-th modal data; Determine the effective index set for the future l-step size of the phase space based on the number of constructible phase space state vectors. The set of valid indexes Need to meet and Count the number of elements in the set of valid indices. ; Calculate the average divergence distance of the nearest neighbor phase space trajectory after the next number of phase space steps based on the nearest neighbor index and the effective index set:

[0032] in To prevent zero from being a very small positive number, the optimal value is selected. , Let be the average logarithmic divergence distance of the j-th mode under the future l-th step size in the k-th window; It should be noted that the phase space step size l has a divergence fitting range of 1- , To maximize the step size for divergent fitting, a step size of 20 is preferred. The uniform sampling frequency is 4Hz. 20 steps correspond to a local evolution time of about 5 seconds, which is sufficient to observe the short-term nonlinear divergent trends of heart rate regulation, respiratory rhythm, skin conductance response and EEG envelope changes, while avoiding excessively long time spans that could lead to non-stationarity of health status affecting the fitting results. Relating the mean divergence distance to time By performing linear fitting, the local trajectory divergence intensity is obtained:

[0033] in, Let be the local trajectory divergence intensity of the j-th modal data in the k-th window. Let l be the actual time corresponding to the l-th step length. Indicates from arrive The mean, Indicates from arrive The mean, To prevent zero from being the smallest positive number; Local trajectory divergence intensity is used to characterize the responsiveness of a user's physiological regulatory system to minor state disturbances. Under healthy conditions, human heart rate, respiration, skin conductance, and brain electrical rhythms typically exhibit moderate nonlinear divergence capabilities. When a user is fatigued, experiencing stress imbalance, or in a state of regulatory rigidity, the trajectory divergence intensity may decrease or exhibit abnormal changes. From the window of users identified as having a healthy status over the past 14 days, the latency and embedding dimension are calculated to construct the collection of the user's historical phase space trajectory. For each phase space state vector in the historical phase space trajectory set, find the nearest neighbor distance:

[0034] in Let be the nearest neighbor distance of the m-th phase space state vector in the historical phase space trajectory set. and These are the m-th and n-th phase space state vectors, respectively; The individual baseline neighborhood radius is obtained by calculating the median distance of the nearest neighbor distances of all phase space state vectors in the historical phase space trajectory set and multiplying it by a neighborhood magnification factor. The neighborhood magnification factor is preferably set to 1.5, which can moderately expand the neighborhood based on the typical nearest neighbor scale of the user's historical stable state, so that the trajectory distribution statistics of the current window have sufficient sample support. For each phase space state vector in the current window's phase space trajectory, the number of neighbor occupancy is counted and normalized to a neighborhood distribution probability. The neighborhood occupancy number represents the number of other phase space state vectors within the individual baseline neighborhood radius of each phase space state vector. The phase space neighborhood distribution entropy is then calculated based on the neighborhood distribution probability.

[0035] in Let the phase space neighborhood distribution entropy of the j-th modal data in the k-th window be denoted as . Let be the neighborhood distribution probability of the i-th phase space vector; Phase space neighborhood distribution entropy is used to characterize the richness of the distribution of a user's physiological state trajectory in phase space. If the trajectory is concentrated in a few regions for a long time, it indicates that the health regulation mode is simple and there may be fatigue or limited physiological regulation. If the trajectory distribution is too discrete, it may also indicate stress disturbance or abnormal fluctuation. Calculate the trajectory velocity between adjacent phase space state vectors based on the current window's phase space trajectory set, and calculate the trajectory energy activity based on the trajectory velocity:

[0036]

[0037] in This represents the trajectory velocity of the j-th modal data at the i-th phase space step in the k-th window. The trajectory energy activity of the j-th modal data in the k-th window; Trajectory energy activity is used to characterize the intensity of a user's dynamic health state as it moves through phase space over time.

[0038] S2.2. Based on any one of the indicators of local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity, establish a robust baseline within the historical health window using the median and median absolute deviation, including the individual baseline center value and the individual baseline stability scale. Robust deviation values ​​are obtained by robustly normalizing the local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity based on a robust baseline. These deviation values ​​include the deviation of the local trajectory divergence intensity of the j-th mode relative to the individual baseline. The deviation of the phase space neighborhood distribution entropy from the individual baseline and the deviation of trajectory energy activity from the individual baseline ; S2.3. To map different indicators to a unified 0-100 vitality scale, an S-shaped mapping function is used to map robustness deviation values ​​to obtain vitality components:

[0039] in The mapping vitality component for the 0th index includes the local trajectory divergence vitality component, the neighborhood distribution entropy vitality component, and the trajectory energy vitality component. The deviation value of the o-th indicator. Let be the mapping sensitivity coefficient of the 0th index; Specifically, for the local trajectory divergence vitality component, its mapping sensitivity coefficient is preferably 0.85, the sensitivity coefficient for the neighborhood distribution entropy vitality component is preferably 0.75, and the sensitivity coefficient for the trajectory energy vitality component is preferably 0.65. The trajectory divergence intensity is more sensitive to changes in physiological regulatory agility and should be given a stronger response in vitality evaluation. Entropy value reflects complex adaptation, but the changes are usually smoother. Energy activity is more easily affected by short-term movement or posture changes, so its mapping sensitivity needs to be reduced. The single-mode chaotic vitality value is obtained by weighting all vitality components. .

[0040] S3. Form a chaotic vitality sequence from the single-modal chaotic vitality value and use it as a time-delayed causal network node sequence. Establish the node self-history state based on the node sequence. Select node pairs to establish the source node lag state and the target node next window state. Combine the node self-history state to establish triplet and analyze the node pairs to generate time-delayed causal edges. S3.1. Starting from the current window k, extract forward. A network construction window is formed by a series of chaotic vitality data windows. In the network construction window, the single-mode chaotic vitality values ​​of each mode are formed into a chaotic vitality sequence, and the chaotic vitality sequence is used as the node sequence of the time-delay causal network, with each mode representing a type of node; To make the chaotic vitality sequences of different users and different modalities comparable, each chaotic vitality value is converted into a standard deviation relative to the user's own health baseline. Convert the node sequence into a tri-state health regulation symbol sequence:

[0041] in This represents the health adjustment symbol of the j-th modal node in the t-th window. Let be the value of the j-th modality node in the t-th window, where 0 represents a low-vitality state below the individual's health baseline, 1 represents a stable state close to the individual's health baseline, and 2 represents a high-vitality state above the individual's health baseline. The symbolic threshold is preferably 0.8, which indicates that the current modality deviates from the individual's health baseline by a robust scale, balancing sensitivity and noise resistance. S3.2, Select the candidate delay set by choosing the largest candidate delay step. The preferred maximum candidate time delay step is 6, and the length of each candidate time delay step is the window length of 300s, that is, the length of the maximum candidate time delay is 30 minutes. The health regulation and conduction between heart rate variability, skin conductance, respiration and brain waves usually occur on the order of minutes to tens of minutes. If the maximum time delay is too short, it is easy to miss the relationship between stress conduction and fatigue accumulation. If the maximum time delay is too long, it is easy to mistakenly connect changes under different activity scenarios as causal relationships. A second-order self-history state is constructed for each node based on the tri-state health regulation symbol sequence:

[0042] in This represents the historical state of the j-th node in the t-th window.

[0043] S3.3, Select any two different nodes to define the source node. and target node Establish the source node's lagging state and the target node's next window state respectively:

[0044] in Time delay The source node is in a lagging state. The next window state for the target node; Based on the source node's lag state, the target node's next window state, and the target node's second-order self-history state, a ternary state is constructed within the network construction window. The frequency of each ternary state is then counted within the network construction window.

[0045] in This indicates that the target future state is in the k-th network construction window. The source lag state is The target is from the historical state. Sample count, , This indicates the number of samples that meet the criteria. Given a time delay The following is a set of valid window indexes that can be used for statistics; Specifically, Represented as:

[0046] Calculate the conditional probability of the target node's next state with the source node participating, and the conditional probability of the target node's next state without considering the source node, based on the number of occurrences of the ternary state:

[0047]

[0048] in This represents the conditional probability given that the source node is involved. This refers to the conditional probability of the next state of the target node without considering the source node. For the target future state, The smoothing coefficient is preferably set to 0.5, which can balance stability and sensitivity in short-window health data. Further statistical analysis of the joint probability of the occurrence of ternary states within the current network construction window:

[0049] in Let be the joint probability of the ternary state within the current network construction window. The source node is in a lagging state. For the target node's historical state, For all possible combinations of states, and There are three possible values. It consists of two tri-state health adjustment symbols of the node, therefore it has 9 possible values. There are 3×3×9 possible values; Calculate the source node based on conditional probability and joint probability. In time lag Next target node Conditional transition entropy:

[0050] in, For the source node In time lag Next target node The conditional transition entropy; S3.4 Calculate the source node based on the node sequence To the target node In time lag Consistency of the following lagged correlation directions:

[0051] in This refers to the directional consistency between the lagging chaotic vitality deviation of the source node and the future chaotic vitality deviation of the target node. express In the effective index set The mean within, express In the effective index set The mean within; Candidate causal edge symbols are determined based on directional consistency, including , And the three symbols: 0

[0052] in Candidate causal edge symbols, This indicates that the target node tends to rise after the source node rises, which is a unidirectional activation edge. This indicates that the target node tends to decrease after the source node increases, which is a reverse suppression edge. 0 indicates that the direction is unstable and no interpretable symbolic edge is formed. The preferred value for the directional consistency threshold is 0.15, which preserves early directional information in the short-window healthy sequence while filtering out weakly correlated fluctuations that have no directional explanatory value. S3.5. Calculate the evidence value of candidate causal edges by combining conditional transition entropy and directional consistency:

[0053] in Indicates the source node To the target node In time lag The candidate causal edge evidence value, The value of 0.3 is preferred as the directional consistency enhancement coefficient. The nonlinear conditional transfer entropy is used as the main causal evidence, while the directionally stable candidate edges are moderately enhanced. For each pair of nodes arrive The time lag with the largest candidate causal edge evidence value is selected as... arrive Optimal causal time delay The system extracts the corresponding maximum candidate causal edge evidence value, calculates the historical maximum candidate causal evidence value set based on the user's historical health status window sequence, extracts the median and median absolute deviation from the historical maximum candidate causal evidence value set, and weights them to obtain the individualized edge evidence threshold. If the maximum candidate causal edge evidence value is greater than the individualized edge evidence threshold and the candidate causal edge sign is not 0, then a node is generated. arrive Determine the time-delayed causal edges and establish a set of causal edges. ; Calculate the final edge weight of the time-delay causal edge:

[0054] in For nodes arrive The final edge weight, The maximum candidate causal edge evidence value. For individualized edge evidence thresholds, To prevent zero from being a very small positive number.

[0055] S4. Establish a causal adjacency matrix based on time-delayed causal edges and calculate the basic health score and short-term deterioration trend penalty term. Combine the basic health score and short-term deterioration trend penalty term to calculate the causal chaotic health index. Based on the causal chaotic health index, identify the user's health status and generate an intervention trigger marker. S4.1 Calculate the causal edge strength for each causal edge in the causal edge set and construct the time-delay causal adjacency matrix:

[0056]

[0057] in For nodes Pointing to node The causal edge strength, The candidate causal edge symbol under optimal causal time delay. It is a time-delayed causal adjacency matrix; S4.2 Calculate the causal input flow and causal output flow of each node based on the time-delay causal adjacency matrix, and subtract the causal output flow from the causal input flow to obtain the net causal flow of the node:

[0058]

[0059] in For modal data, For nodes Total causal output strength for other modalities Indicates from node arrive The absolute value of the causal edge strength, For nodes The total causal input strength influenced by other modal impressions, Represents a node arrive The absolute value of the causal edge strength; Calculate the network causal density and average causal propagation strength based on the causal edge set:

[0060]

[0061] in For network causal density, For the average causal transmission strength, There are 4 nodes, representing four different data modalities. Indicates the number of valid causal edges; The causal network load is obtained by weighting the network causal density and the average causal propagation strength. Calculate the causal source concentration based on the net causal flow:

[0062] in For the purification of cause and effect, This indicates that only the positive net output portion is taken. Indicates the concentration of causal sources; Weighting the causal source concentration and causal network load and truncating the result in a causal network anomaly risk within the range of 0 to 1, and subtracting the causal network anomaly risk from 1, yields the causal network structural health component. ; S4.3. Weight all single-modal chaotic vitality values ​​to obtain the total chaotic vitality value. Extract the set of historical total chaotic vitality values ​​from the user's historical health status window, and calculate the median and median absolute deviation. Normalize and truncate the total chaotic vitality value using the median and median absolute deviation to obtain the risk of insufficient chaotic vitality.

[0063] in Risk of insufficient chaotic vitality The median. This represents the absolute deviation of the median. The total chaotic vitality value is represented by `clip`, which is a truncation operation that restricts the value to between 0 and 1. The chaotic vitality health component is obtained by subtracting the risk of insufficient chaotic vitality from 1. The health component of the causal network structure And Chaos Vitality Health Quantity The basic health score is obtained by weighted summation. ; S4.4, Define the number of trend review periods :

[0064] in The preferred duration for trend review is 10 minutes. Short-term stress activation, decreased attention, fatigue accumulation, or abnormal motor recovery typically form observable trends within 5 to 15 minutes. Choosing 10 minutes balances sensitivity and noise resistance. The window time interval is preferably 5 minutes; Calculate the decline rate of the chaotic vitality health component and the decline rate of the causal network structure health component based on the number of trend retrospective cycles:

[0065]

[0066] in The extent of the decline in the chaotic vitality and health component. The extent of the decline in the health component of the causal network structure; The short-term deterioration trend penalty term is obtained by weighted summing the decline rates of the chaotic vitality health component and the causal network structure health component, and then truncating the sum within the range of 0 to 1. The causal chaotic health index is then calculated based on the short-term deterioration trend penalty term.

[0067] in For causal chaos health index, The trend penalty intensity coefficient is preferably set to 0.25, which means that when the short-term deterioration trend reaches its maximum, the basic health score will be reduced by a maximum of 25%, which can reflect the risk of rapid deterioration without allowing the trend item to completely replace the current true health level.

[0068] S4.5. Identify user health status based on causal chaotic health index, including green stable state, yellow compensated state, and red decompensated state. If the causal chaotic health index is greater than or equal to the yellow warning threshold, it is a green stable state. If it is less than the yellow warning threshold but greater than or equal to the red warning threshold, it is a yellow compensated state. If it is less than the red warning threshold, it is a red decompensated state. The yellow and red warning thresholds are determined by the causal chaotic health index in the user health status window. The yellow and red warning thresholds are taken as the 78th percentile and 52nd percentile, respectively. The yellow warning threshold indicates that the user has shown a decline in chaotic vitality, enhanced causal transmission, or a short-term deterioration trend. The red warning threshold indicates that the user's current regulatory system has deviated significantly from the individual health baseline. An intervention trigger marker is generated for each user's health status, including 0, 1, and 2, corresponding to the green stable state, the yellow compensated state, and the red decompensated state, respectively. Corresponding intervention measures are taken for each intervention trigger marker. Specifically, no intervention is taken for the green stable state, and only the user's health data is recorded. For the yellow compensated state, a light reminder is pushed to the user's terminal, such as a rest reminder, breathing regulation suggestions, or exercise intensity prompts. For the red decompensated state, a strong intervention inference is triggered, such as immediately stopping high-intensity activities, prompting manual attention, or proceeding to step six of the counterfactual intervention inference.

[0069] This embodiment also provides a user health monitoring and management system based on big data, including: The raw data analysis module is used to collect user multimodal physiological data, preprocess and resample it to form a health analysis sequence, and construct a phase space state vector based on the analysis of the original candidate embedding dimension of the health analysis sequence. The Chaotic Vitality Mapping Module is used to calculate spatial indices based on phase space state vectors, including local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity. It establishes a robust baseline based on the spatial indices and maps them to obtain the single-mode chaotic vitality value. The causal network construction module is used to form a chaotic vitality sequence from the single-modal chaotic vitality value and use it as a time-delayed causal network node sequence. Based on the node sequence, the self-historical state of the node is established. The source node's lagging state and the target node's next window state are selected by node pairs. Triples are established by combining the node's self-historical state and the node pairs are analyzed to generate time-delayed causal edges. The health analysis module is used to establish a causal adjacency matrix based on time-delayed causal edges and calculate a basic health score and a short-term deterioration trend penalty term. It then calculates a causal chaotic health index by combining the basic health score and the short-term deterioration trend penalty term, identifies the user's health status based on the causal chaotic health index, and generates an intervention trigger marker.

[0070] In summary, this invention collects multimodal physiological data, performs preprocessing and resampling, constructs a unified health analysis sequence, and generates state vectors based on phase space reconstruction, thereby effectively expressing the nonlinear dynamic characteristics of complex physiological systems and enhancing the structured representation capability of raw time-series data. Furthermore, by calculating local trajectory divergence intensity, neighborhood entropy, and trajectory energy activity and establishing a robust baseline, a single-modal chaotic vitality value is obtained, normalizing individual differences and improving cross-user comparability. Then, the chaotic vitality sequence is used as a time-delayed causal network node, introducing a lagged state and triplet modeling approach to characterize the temporal causal propagation relationship between multimodalities. Finally, a causal chaotic health index is constructed by combining a causal adjacency matrix with a basic health score and a short-term deterioration penalty term, enabling dynamic quantitative assessment, risk classification, and intervention triggering of health status, thus significantly improving the accuracy, temporal sensitivity, and interpretability of health monitoring.

[0071] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A user health monitoring and management method based on big data, characterized in that: include, Multimodal physiological data of users are collected, preprocessed and resampled to form a health analysis sequence, and the original candidate embedding dimension is determined based on the health analysis sequence to construct the phase space state vector; Spatial indices, including local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity, are calculated based on phase space state vectors. A robust baseline is established based on the spatial indices and mapped to obtain the single-mode chaotic activity value. The single-modal chaotic vitality values ​​are formed into a chaotic vitality sequence and used as a time-delayed causal network node sequence. Based on the node sequence, the node self-historical state is established. Node pairs are selected to establish the source node's lag state and the target node's next window state. Triples are established by combining the node self-historical state and the node pairs are analyzed to generate time-delayed causal edges. A causal adjacency matrix is ​​established based on time-delayed causal edges, and a basic health score and a short-term deterioration trend penalty term are calculated. A causal chaotic health index is calculated by combining the basic health score and the short-term deterioration trend penalty term. The user's health status is identified based on the causal chaotic health index, and an intervention trigger marker is generated.

2. The user health monitoring and management method based on big data as described in claim 1, characterized in that: The process of collecting user multimodal physiological data, preprocessing it, and resampling it to form a health analysis sequence specifically involves: collecting raw user multimodal physiological data through intelligent monitoring devices. It includes four types of physiological modal data: heart rate variability modality (h), skin conductance modality (s), respiratory modality (r), and electroencephalogram (EEG) modality (b). Define the length of the health analysis window and update step size Define the k-th health analysis window ; The raw multimodal physiological data falling into the health analysis window are extracted and preprocessed for each modality. The preprocessed multimodal physiological data were uniformly resampled to the health standard time axis to obtain resampled standard modality data. And form a standardized multimodal health analysis sequence. .

3. The user health monitoring and management method based on big data as described in claim 2, characterized in that: The specific steps for determining the original candidate embedding dimensions and constructing the phase space state vector based on the health analysis sequence are as follows: for the j-th class of standard modal data sequence... Slow-varying trend terms are obtained by using fixed-length median trend filtering; From standard modal data sequences Subtracting the slow-changing trend term from the middle yields the health dynamic sequence. ; The health dynamic sequence is divided into equal-frequency binning methods. A range of amplitude values ​​is defined, and candidate delays are defined based on these ranges. Calculate the average mutual information of candidate delays; The first local minimum is selected as the candidate delay time based on the average mutual information. ; Define candidate embedding dimensions And construct a temporary phase space vector based on the candidate embedding dimension; For each temporary phase space vector, find the nearest neighbor index. The proportion of false nearest neighbors is calculated based on the nearest neighbor index and the temporary phase space vector. Determine the original candidate embedding dimension based on the proportion of false nearest neighbors; Construct the phase space state vector using the original candidate embedding dimension. The phase space state vectors of all phase spaces are combined to form the set of phase space trajectories of the j-th mode data. .

4. The user health monitoring and management method based on big data as described in claim 3, characterized in that: The calculation of spatial indices based on phase space state vectors specifically involves: determining the number of constructible phase space state vectors based on the original candidate embedding dimensions; Determine the effective index set for the future l-step size of the phase space based on the number of constructible phase space state vectors. ; Calculate the average divergence distance of the phase space trajectory of the nearest neighbor point after the next l phase space steps based on the nearest neighbor index and the effective index set; The local trajectory divergence intensity is obtained by linearly fitting the average divergence distance; Constructing the collection of user historical phase space trajectories For each phase space state vector in the historical phase space trajectory set, find the nearest neighbor distance; The individual baseline neighborhood radius is obtained by calculating the median distance of the nearest neighbor distances of all phase space state vectors in the historical phase space trajectory set and multiplying it by a neighborhood magnification factor. ; For each phase space state vector in the current window phase space trajectory, the number of neighborhood occupancy is counted and normalized to the neighborhood distribution probability. The phase space neighborhood distribution entropy is calculated based on the neighborhood distribution probability. The trajectory velocity between adjacent phase space state vectors is calculated based on the current window's phase space trajectory set, and the trajectory energy activity is calculated based on the trajectory velocity.

5. The user health monitoring and management method based on big data as described in claim 4, characterized in that: The specific steps for establishing a robust baseline based on spatial indicators and mapping it to obtain the single-modal chaotic vitality value are as follows: a robust baseline is established within a historical health window using the median and median absolute deviation based on any one of the indicators among local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity. Based on the robust baseline, robust deviation values ​​are obtained by robustly normalizing the local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity. The vitality component is obtained by mapping the robust deviation value using an S-shaped mapping function; The single-mode chaotic vitality value is obtained by weighting all vitality components. .

6. The user health monitoring and management method based on big data as described in claim 5, characterized in that: The process of forming a chaotic vitality sequence from single-modal chaotic vitality values ​​and using it as a time-delay causal network node sequence, and establishing the node's self-historical state based on the node sequence, specifically involves: forming a network construction window with the current window k as the endpoint. In the network construction window, the single-mode chaotic vitality values ​​of each mode are formed into a chaotic vitality sequence, and the chaotic vitality sequence is used as the node sequence of the time-delay causal network. The node sequence is converted into a three-state health regulation symbol sequence, and a second-order self-history state is constructed for each node based on the three-state health regulation symbol sequence.

7. The user health monitoring and management method based on big data as described in claim 6, characterized in that: The process of selecting nodes to establish the source node's lag state and the target node's next-window state, combining the nodes' historical states to create triplet pairs, and analyzing the generation of time-delayed causal edges between node pairs specifically involves: selecting any two different nodes to define the source node. and target node Establish the source node's lag state and the target node's next window state respectively; Based on the source node's lag state, the target node's next window state, and the target node's second-order self-historical state, a window is constructed to count the ternary states, and the number of occurrences of each ternary state is counted in the network construction window. Calculate the conditional probability of the target node's next state under the condition that the source node participates and the conditional probability of the target node's next state without considering the source node based on the number of occurrences of the ternary state. Further, calculate the joint probability of the number of occurrences of the ternary state in the current network construction window. Calculate the source node based on conditional probability and joint probability. In time lag Next target node The conditional transition entropy; Calculate the source node based on the node sequence. To the target node In time lag Consistency of the lagged correlation direction; Candidate causal edge symbols are determined based on directional consistency, including three symbols: +1, -1, and 0. The evidence value of candidate causal edges is calculated by combining conditional transition entropy and directional consistency. For each pair of nodes arrive The time lag with the largest candidate causal edge evidence value is selected as... arrive Optimal causal time delay The system extracts the corresponding maximum candidate causal edge evidence value, calculates the historical maximum candidate causal evidence value set, and extracts the median and median absolute deviation from the historical maximum candidate causal evidence value set for weighted summation to obtain the individualized edge evidence threshold. If the maximum candidate causal edge evidence value is greater than the individualized edge evidence threshold and the candidate causal edge sign is not 0, then a node is generated. arrive Determine the time-delayed causal edges and establish a set of causal edges. ; Calculate the final edge weight of the time-delayed causal edge.

8. The user health monitoring and management method based on big data as described in claim 7, characterized in that: The specific steps for establishing a causal adjacency matrix based on time-delayed causal edges and calculating the basic health score and short-term deterioration trend penalty term, and then calculating the causal chaotic health index by combining the basic health score and the short-term deterioration trend penalty term, are as follows: calculate the causal edge strength based on each causal edge in the causal edge set and construct a time-delayed causal adjacency matrix; The causal input flow and causal output flow of a node are calculated based on the time-delay causal adjacency matrix, and the net causal flow of the node is obtained by subtracting the causal output flow from the causal input flow. Calculate the causal density and average causal propagation strength of the network based on the causal edge set; The causal network load is obtained by weighting the network causal density and the average causal propagation strength. Calculate the causal source concentration based on the net causal flow; Weighting the causal source concentration and causal network load and truncating the result in a causal network anomaly risk within the range of 0 to 1, and subtracting the causal network anomaly risk from 1, yields the causal network structural health component. ; The total chaos vitality value is obtained by weighting all single-mode chaos vitality values. The total chaos vitality value is then normalized and truncated to obtain the risk of insufficient chaos vitality. The chaotic vitality health component is obtained by subtracting the risk of insufficient chaotic vitality from 1. The health component of the causal network structure And Chaos Vitality Health Quantity The basic health score is obtained by weighted summation. ; Define the number of trend retracing periods The decline rate of the chaotic vitality health component and the decline rate of the causal network structure health component were calculated based on the number of trend review cycles. The short-term deterioration trend penalty term is obtained by weighted summing of the decline in the chaotic vitality health component and the decline in the causal network structure health component and truncating it in the range of 0 to 1. The causal chaotic health index is calculated based on the short-term deterioration trend penalty term.

9. The user health monitoring and management method based on big data as described in claim 8, characterized in that: The specific steps for identifying user health status and generating intervention trigger markers based on the causal chaotic health index are as follows: identifying user health status based on the causal chaotic health index, including green stable status, yellow compensated status, and red decompensated status; An intervention trigger marker is generated for each user's health status, including 0, 1, and 2, which correspond to the green stable status, the yellow compensated status, and the red decompensated status, respectively. Corresponding intervention measures are taken for each intervention trigger marker.

10. A user health monitoring and management system based on big data, based on the user health monitoring and management method based on big data as described in any one of claims 1 to 9, characterized in that: include, The raw data analysis module is used to collect user multimodal physiological data, preprocess and resample it to form a health analysis sequence, and construct a phase space state vector based on the analysis of the original candidate embedding dimension of the health analysis sequence. The Chaotic Vitality Mapping Module is used to calculate spatial indices based on phase space state vectors, including local trajectory divergence intensity, phase space neighborhood distribution entropy, and trajectory energy activity. It establishes a robust baseline based on the spatial indices and maps them to obtain the single-mode chaotic vitality value. The causal network construction module is used to form a chaotic vitality sequence from the single-modal chaotic vitality value and use it as a time-delayed causal network node sequence. Based on the node sequence, the self-historical state of the node is established. The source node's lagging state and the target node's next window state are selected by node pairs. Triples are established by combining the node's self-historical state and the node pairs are analyzed to generate time-delayed causal edges. The health analysis module is used to establish a causal adjacency matrix based on time-delayed causal edges and calculate a basic health score and a short-term deterioration trend penalty term. It then calculates a causal chaotic health index by combining the basic health score and the short-term deterioration trend penalty term, identifies the user's health status based on the causal chaotic health index, and generates an intervention trigger marker.

Images