Health information monitoring and management system based on multi-source data fusion analysis
A health information monitoring system that integrates and analyzes multi-source data solves the problem of biased health status assessment caused by a single data source, enabling dynamic assessment and personalized intervention of individual health status, and is suitable for health management in high-risk environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- BEIJING YIAN MEDICAL TECHNOLOGY CO LTD
- Filing Date
- 2026-01-22
- Publication Date
- 2026-05-01
AI Technical Summary
Existing health information monitoring systems mostly rely on a single data source and lack the ability to integrate and deeply analyze multi-source heterogeneous data. This makes it easy for individual health status assessment results to be biased and difficult to identify potential health risks. In particular, under high-pressure, fast-paced, and high-risk environments, there is a lack of continuous perception and forward-looking intervention mechanisms for subtle changes in characteristics.
The health information monitoring and management system, which employs multi-source data fusion analysis, constructs a multi-dimensional feature system through modules for data collection, preprocessing, feature fusion, health status assessment, risk detection, and intervention suggestion generation. This system combines non-negative tensor decomposition and counterfactual inference methods to achieve dynamic assessment and intervention suggestions for individual health status.
It significantly improves the accuracy and individualized identification capabilities of health information monitoring, and can identify the combined effects of physiological signals, behavioral characteristics and environmental factors in the time-series dimension. It can dynamically assess the combined stress load and sleep repair efficiency, and generate intervention recommendations that are interpretable, targeted and practical, and is suitable for remote health management and high-risk work environments.
Smart Images
Figure CN121964145A_ABST
Abstract
Description
Health Information Monitoring and Management System Based on Multi-Source Data Fusion Analysis Technical Field
[0001] This invention relates to the field of health data collection and monitoring technology, specifically to a health information monitoring and management system based on multi-source data fusion analysis. Background Technology
[0002] In recent years, with the rapid development of mobile medical devices, wearable sensors, and remote monitoring technologies, the means of acquiring personal health data have become increasingly diversified, covering multiple dimensions such as heart rate, blood oxygen, body temperature, exercise trajectory, diet, and sleep. However, existing health information monitoring systems mostly rely on a single data source for status assessment, lacking the ability to fuse and deeply analyze multi-source heterogeneous data. This leads to biased individual health status assessment results and makes it difficult to accurately identify potential health risks.
[0003] Especially in high-pressure, fast-paced, and unstable work environments (such as high-altitude construction sites, ocean-going ship control rooms, and underground mines), an individual's health status may change drastically in a short period of time. For example, in the low-oxygen environment of high altitudes, blood oxygen data alone is insufficient to fully reflect the risk of altitude sickness; dynamic analysis combining multiple indicators such as heart rate variability, sleep structure, and activity patterns is necessary.
[0004] Furthermore, traditional health management methods are mostly reactive, intervening only after a user experiences obvious physiological abnormalities, lacking continuous perception and proactive intervention mechanisms for subtle changes in characteristics. Therefore, there is an urgent need for a health information monitoring and management method based on multi-source data fusion analysis to achieve early warning, proactive intervention, and personalized management, especially in emergency and high-risk work environments, providing users with continuous, dynamic, and accurate health protection. Summary of the Invention
[0005] The purpose of this invention is to provide a health information monitoring and management system based on multi-source data fusion analysis to address the shortcomings of the prior art.
[0006] To achieve the above objectives, the present invention provides the following technical solution: a health information monitoring and management system based on multi-source data fusion analysis, comprising: a data acquisition module for acquiring multi-source raw health data of a preset object within a preset time period, wherein the multi-source raw health data includes a first type of physiological signal data, a second type of behavioral characteristic data, and a third type of environmental exposure data; a data preprocessing module for standardizing the first, second, and third types of data respectively to obtain corresponding normalized feature sequences S1, S2, and S3; and a feature fusion module for constructing a multi-dimensional fusion feature vector group F, wherein each feature is composed of normalized S1, S2, and S3. The data analysis and calculations under the corresponding time windows in the three modules are as follows: The health status assessment module, based on a multi-dimensional fusion feature vector group F, inputs the trained health status classification model and outputs the health status level label L corresponding to each time period; the risk detection module, based on continuous health status level labels L, dynamically tracks regions where health status changes abruptly in the time series, identifying potential health risk nodes P={p1,p2,...,pj,...,pm}, where each pj represents a mutation start point or trend turning point; the intervention suggestion generation module generates intervention suggestions based on the changing trend of the fusion features corresponding to the health risk node P.
[0007] Preferably, the first type of physiological signal data includes heart rate, blood oxygen saturation, and respiratory rate data.
[0008] Preferably, the second type of behavioral characteristic data includes steps, exercise intensity, sleep structure, and number of times the body turns over.
[0009] Preferably, the third type of environmental exposure data includes noise intensity, ambient temperature and humidity, and atmospheric pressure.
[0010] Preferably, the data preprocessing module includes data cleaning, missing value interpolation, time alignment, and standardization. The standardization includes normalizing each feature value to the [0,1] interval.
[0011] Preferably, the feature fusion module performs non-negative tensor decomposition on the normalized feature sequence by constructing a tensor structure, extracts latent stress factors, and calculates the composite stress load index for each time period as part of the multidimensional fusion feature vector group F.
[0012] Preferably, the feature fusion module further includes constructing a counterfactual sleep scenario based on sleep structure and pre-existing physiological state, combining it with a physiological recovery prediction model, calculating a sleep repair efficiency index for each time period, and using it as part of the multidimensional fusion feature vector group F.
[0013] Preferably, the health status assessment module uses a random forest classification model constructed with a supervised learning algorithm. Based on the composite stress load index and sleep repair efficiency index, it outputs the health status level for the corresponding time period. The health status level includes three types: normal, sub-healthy, and abnormal warning.
[0014] Preferably, the risk detection module performs sliding window difference analysis on the health status level marker sequence, and identifies the location where the health status level changes abruptly as a potential health risk node by combining the set thresholds of the state level transition direction and fluctuation amplitude.
[0015] Preferably, the intervention suggestion generation module analyzes the trend of changes in the integrated characteristics of the time period before and after the health risk node, identifies the characteristic patterns of abnormal growth, abnormal decline or continuous deviation from the normal range, and generates intervention suggestions containing suggestion type, suggestion content and triggering reason explanation according to a predefined intervention rule base.
[0016] In the above technical solution, the technical effects and advantages provided by this invention are as follows: 1. By collecting and fusing multi-source health data, and combining advanced computational methods such as nonnegative tensor decomposition and counterfactual inference, this invention constructs a multi-dimensional feature system that comprehensively reflects an individual's health status, significantly improving the accuracy of health information monitoring and individualized identification capabilities. The system can not only identify the combined influence of physiological signals, behavioral characteristics, and environmental factors in the temporal dimension, but also dynamically assess compound stress load and sleep repair efficiency, thereby achieving a more scientific health status assessment.
[0017] 2. This invention, by introducing a risk mutation detection and intervention suggestion generation mechanism, can generate interpretable, targeted, and practical intervention suggestion texts based on a feature change trend matching intervention strategy rule base after identifying key nodes of abnormal health status trends. Compared to the limitations of existing technologies that rely on static thresholds and single data sources, this invention has the comprehensive advantages of full-cycle dynamic perception, model-driven classification, and trend-aware intervention, making it particularly suitable for scenarios such as remote health management and health monitoring in high-risk work environments. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0019] Figure 1 is a flowchart of the system modules of the present invention. Detailed Implementation
[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0021] As shown in Figure 1, the health information monitoring and management system based on multi-source data fusion analysis described in this embodiment includes: a data acquisition module, which collects multi-source raw health data of a preset object within a preset time period. The multi-source raw health data includes a first type of physiological signal data, a second type of behavioral characteristic data, and a third type of environmental exposure data. In this embodiment, multi-source raw health data is collected from the target monitoring object within the preset time period. The multi-source raw health data includes the following three types: First type: physiological signal data; physiological signal data is collected through wearable devices (such as smart bracelets, chest patch health monitors, wristband physiological parameter recorders, etc.). The collected parameters include, but are not limited to: heart rate (HeartRate): continuously collected at a second-level sampling frequency to reflect the current cardiovascular activity status; blood oxygen saturation ( ): Used to assess blood oxygenation capacity, especially suitable for high-altitude and low-oxygen environments; Respiratory Rate: Used to help identify fatigue, stress, or potential respiratory problems.
[0022] The second category is behavioral characteristic data. Behavioral data is mainly collected through inertial sensors (such as triaxial accelerometers and gyroscopes) and sleep monitoring modules. Parameters include: steps and exercise intensity: used to assess activity levels and daily behavior patterns; sleep duration and structure (such as deep sleep / light sleep / wake stages): reflecting recovery status and circadian rhythm health; number of turns and frequency of body position changes: used to reflect sleep quality and abnormal nighttime behaviors.
[0023] The third category: Environmental exposure data; Environmental data is acquired through external sensor modules or environmental sensing elements built into mobile terminals, mainly including: Ambient noise intensity (dB): High noise environments may affect sleep and stress levels; Ambient temperature and humidity: Related to body temperature regulation, comfort, and heart rate changes; Atmospheric pressure (especially applicable to high-altitude work scenarios): Affects oxygenation capacity.
[0024] For example, when deploying this system at a high-altitude scientific research site, each staff member is issued a smart bracelet integrating multiple sensors such as heart rate, blood oxygen, acceleration, temperature, and humidity. Environmental monitoring equipment is also deployed in dormitories and work areas. The system samples data every minute and aggregates it to an edge computing terminal via Bluetooth or LoRa network, then periodically uploads it to a health monitoring platform.
[0025] Through the above methods, the system can continuously, with low power consumption and high precision collect multi-source raw health data of individuals without interfering with their daily activities, providing basic data support for subsequent data fusion, status assessment and intervention recommendations.
[0026] The data preprocessing module performs standardization processing on the first, second, and third types of data respectively to obtain the corresponding normalized feature sequences S1, S2, and S3.
[0027] In this embodiment, the data preprocessing module is used to standardize the first type of physiological signal data, the second type of behavioral characteristic data, and the third type of environmental exposure data collected from multiple sources, in order to eliminate the data heterogeneity between different units, different sampling frequencies, and different sources, and to ensure the accuracy and consistency of subsequent fusion analysis.
[0028] The data preprocessing module includes, but is not limited to, the following steps: Data cleaning: removing obvious outliers and missing data; Data interpolation: filling in missing sampling points with linear interpolation or moving average; Time synchronization: aligning data of different frequencies to a unified time axis through interpolation or resampling; Standardization: mapping data to [0,1] or [0,1] using normalization methods. Intervals; Output normalized feature sequences: denoted as S1 (physiological category), S2 (behavioral category), and S3 (environmental category), respectively.
[0029] Category 1: Standardization of physiological signal data: Taking heart rate as an example, the raw heart rate value is collected as HR={78,82,110,90,70}. Within the statistical range of the past 30 minutes, the minimum value is set to 60 and the maximum value to 120. Then, the Min-Max normalization method is used to calculate: Similar methods for blood oxygen Processing parameters such as respiratory rate yields a normalized feature sequence S1={s11,s12,...,s1n}.
[0030] Category 2: Standardization of behavioral characteristic data: Taking sleep duration as an example, sleep time is statistically analyzed by hours, with a statistical range of [0,10] hours and an original value of 6.5 hours, then: For behavioral data such as the number of times a person rolls over and the number of steps, which are counted on a daily basis, the Z-score standardization method can be used to obtain a normalized feature sequence S2={s21,s22,...,s2n}.
[0031] Category 3: Standardization of Environmental Exposure Data: Taking ambient temperature as an example, if the human comfort range is set to 18°C to 30°C, and the initial temperature is 26°C, then: For noise data, the threshold range can be set to [30 dB, 90 dB], and the air pressure can be set to the normal physiological influence range [700 hPa, 1050 hPa]. The same normalization method can be used to output the normalized sequence S3={s31,s32,...,s3n}.
[0032] All normalized feature values are generated into sequences using a uniform time window (e.g., 5 minutes, 15 minutes, or 60 minutes), resulting in: normalized feature sequences for physiological signals: S1={s11,s12,...,s1n}; normalized feature sequences for behavioral characteristics: S2={s21,s22,...,s2n}; and normalized feature sequences for environmental exposure: S3={s31,s32,...,s3n}. At each time point (e.g., every 5 minutes), a set of {s1i,s2i,s3i} data is generated for subsequent fusion modeling steps.
[0033] This embodiment significantly improves data consistency and comparability by standardizing and time-aligning health data from different sources and at different scales, laying a solid foundation for subsequent multi-dimensional fusion, model classification, and risk identification. This module has good scalability, making it easy to deploy across different platforms or hardware devices, and is suitable for various scenarios involving dynamic perception of individual health status.
[0034] The feature fusion module constructs a multi-dimensional fusion feature vector group F, where each feature is calculated from data analysis within the corresponding time window in normalized S1, S2, and S3.
[0035] In this embodiment, based on the normalized multi-source data sequences S1 (physiological signals), S2 (behavioral characteristics), and S3 (environmental exposure), a multi-dimensional fusion feature vector group F={f1,f2,...,fi,...,fn} is constructed within each unified time window (e.g., 5 minutes). Each feature fi is calculated jointly from one or more sub-features from S1, S2, and S3, reflecting the dynamic trend and potential risks of an individual's health status. In this embodiment, F includes a composite stress load index and a sleep repair efficiency index.
[0036] The composite stress load index is obtained by constructing a three-dimensional tensor structure from the collected normalized health data. Where: I is the number of feature dimensions (e.g., HRV, ...). (Noise, Temp, etc.); J is the number of time windows (e.g., one window every 5 minutes); K is the number of the monitored object or device (supports multi-person / multi-device modeling); if it is a single-person monitoring system, then K=1, which is simplified to a two-dimensional tensor sequence (multi-feature × time).
[0037] Each tensor element This represents the value of the i-th normalized feature of the k-th object within the j-th time window.
[0038] Example Feature Set (Dimension I): Feature Number (i) Feature Name Source Category 1 HRV' (Heart Rate Variability) Physiological 2 SpO2' (Blood Oxygen) Physiological 3 Resp' (Respiratory Rate) Physiological 4 Noise (Environmental Noise) Environment 5 Temp (Temp) Environment The table uses the CANDECOMP / PARAFAC(CP) model for nonnegative tensor decomposition, with the goal of decomposing the tensor... It can be approximated as the product of factors of a set of low-rank matrices: Where: A∈ : Feature factor matrix (representing the contribution of each feature to each latent dimension), B∈ : Time factor matrix (representing the response intensity of each time period in each latent dimension), C∈ : Individual factor matrix (representing the participation of each object in each latent dimension), R is the decomposition rank (can be between 2 and 5), symbol Represents the outer product of vectors.
[0039] Select each row vector in the time factor matrix B obtained from the NTF decomposition , representing the performance of the j-th time window across different dimensions of latent stress factors. Then, the Composite Stress Load Index (CSLI) for each time period is constructed, with the expression: In the formula, Let be the intensity of the j-th time window on the r-th latent stress dimension; The weight coefficient of the r-th latent factor (can be trained or set based on clinical experience).
[0040] Clinical or health implications of the CSLI index: Low value (<0.3): good physiological and environmental condition, low stress load; moderate value (0.3~0.7): mild stress response or environmental adaptation load; high value (>0.7): possible persistent physiological abnormality or adverse external environment, intervention mechanism is recommended.
[0041] The method for obtaining the sleep repair efficiency index is as follows: A physiological load feature vector is constructed within a preset time window before sleep begins (e.g., 30 minutes before falling asleep). ;in, To normalize heart rate, To normalize heart rate variability, This is a normalized blood oxygen saturation vector; it is used to characterize the "basal state to be repaired." Based on sleep monitoring results, actual sleep structure parameters are extracted: ;in: For deep sleep duration, Total sleep duration The number of nighttime awakenings; without altering the individual's baseline physiological state X, construct at least one counterfactual sleep scenario: Where: Δ is the preset deep sleep enhancement amount (e.g., +15%), δ is the preset arousal reduction amount, and all counterfactual variables satisfy the physiologically reasonable boundary. This counterfactual is used to simulate "what would happen if I slept better". Based on historical samples or training data, a recovery function is established: Where: f is the physiological recovery prediction model (which can be linear regression, nonlinear regression, or neural network), and the physiological recovery amount Y can be defined as: ; calculate separately: The sleep repair efficiency index (SRE) is calculated using the following expression: .
[0042] Where: SRE>0: actual sleep is better than the counterfactual assumption; SRE≈0: sleep repair efficiency is low; SRE<0: sleep structure does not effectively utilize repair potential.
[0043] The health status assessment module, based on a multi-dimensional fusion feature vector group F, takes the trained health status classification model as input and outputs the health status level label L corresponding to each time period.
[0044] In this embodiment, the health status level is classified and determined by combining the composite stress load index and the sleep repair efficiency index. Specifically, the following steps are included: Within each preset time period (e.g., every 30 minutes as a cycle), two key indicators are first extracted from the aforementioned multidimensional fusion feature vector group: Composite stress load index: representing the comprehensive physiological and environmental load level of the individual in the current time period, which has been calculated by the non-negative tensor decomposition method in the aforementioned embodiment; Sleep repair efficiency index: representing the contribution of the individual's previous sleep cycle to the recovery of the current physiological state in this time period, which has been obtained by the counterfactual inference method in the aforementioned embodiment.
[0045] Using these two indicators as the core features of the current time period, a set of binary feature vectors for classification is constructed, represented as: Current time period feature vector = [composite stress load index, sleep repair efficiency index]; the above feature vector will be used as the input of the subsequent health status classification model.
[0046] The health status classification model is a multi-class discriminant model constructed using supervised learning methods, used to predict an individual's current health status level based on the input feature vector. The model construction includes the following specific steps: Training data construction: Based on historical data, training samples with labeled health statuses are collected from multiple time periods. The samples include the composite stress load index, sleep repair efficiency index, and corresponding manually labeled health status levels (e.g., determined by medical experts based on actual examination results); Feature space normalization: The composite stress load index and sleep repair efficiency index are numerically normalized to unify the scale of the input features. Methods such as min-max normalization are used to map feature values to the [0,1] interval; Classification model selection and training: The random forest classification algorithm is selected for model construction. This algorithm has strong feature nonlinear modeling capabilities and anti-overfitting capabilities, making it suitable for health status classification problems with small sample sizes. During training, 5-fold cross-validation is used to evaluate model stability, and the final output is the trained random forest classifier.
[0047] During each inference, the normalized binary feature vector of the current time period is input into the trained classifier to obtain the health status probability output predicted by the model.
[0048] The model returns a three-class probability distribution for each input feature vector, representing the probability value of an individual belonging to each of the following health states in the current time period: normal, sub-healthy, and abnormal warning. Based on the probability results output by the model, the category corresponding to the highest probability is selected as the health state level label for the current time period. For example, if the model output is: normal: 0.25, sub-healthy: 0.40, and abnormal warning: 0.35, then the health state level label for the current time period is "sub-healthy".
[0049] At the same time, a threshold control mechanism is set. For example, when the predicted probability of the "abnormal warning" category is higher than 0.6, regardless of whether it is the maximum probability, the output is forced to be "abnormal warning" to ensure sensitive identification in high-risk situations.
[0050] Because the model output may experience drastic fluctuations in health status levels due to short-term anomalies, potentially resulting in isolated state transitions of the "abnormal-normal-abnormal" type, a smoothing process is performed on the output state level sequence to improve the continuity and medical interpretability of health trends. The specific steps are as follows: Constructing a state time series: Arranging the health status level labels output in continuous time periods in chronological order to form a state time series; Sliding window smoothing algorithm: Setting a fixed-length sliding window (e.g., 3 time periods). If the state level of the central time period is inconsistent with the state levels on both sides, and its duration is less than the length of a smoothing window, then the central state is corrected according to the majority of states within the window; Isolated fluctuation elimination rule: If an abnormal warning state only appears in a single time period, and the preceding and following states are normal or sub-healthy, then the state can be considered a false trigger and eliminated or downgraded according to the threshold stabilization strategy; Outputting the final state sequence: After smoothing, a continuous and highly reliable health status level label sequence is obtained, which is used in the subsequent health intervention suggestion generation step.
[0051] By combining the composite stress load index and sleep repair efficiency index, a trained classification model is used to dynamically identify health status. A state smoothing mechanism is introduced to improve the temporal consistency of the judgment results, thereby enhancing the stability, accuracy and clinical reference value of health monitoring results.
[0052] The risk detection module, based on continuous health status level labels L, dynamically tracks regions where health status changes abruptly in a time series, and identifies potential health risk nodes P={p1,p2,...,pj,...,pm}, where each pj represents a mutation start point or trend inflection point.
[0053] First, obtain the health status level label sequence output by the health status classification model trained in the previous stage. This sequence is arranged in chronological order and is represented as: Health status level sequence L={l1, l2, ..., lj, ..., ln}; where each status level label lj represents the health status level in the j-th time period, and its value comes from a fixed set of levels, for example: normal state: coded as 0; sub-healthy state: coded as 1; abnormal warning state: coded as 2; Convert this status level label sequence into an integer encoding sequence: status encoding sequence E={e1, e2, ..., ej, ..., en}, where ej∈{0,1,2}; this encoding form facilitates subsequent gradient calculation and trend analysis.
[0054] Apply a fixed-length sliding window method to the state coding sequence, setting the window length to w, with a recommended value of 3 or 5 (e.g., w=3), and a sliding step size of 1.
[0055] For each window centered at the j-th time point, extract its state encoding values before and after the time point, and calculate the difference Δe between the state changes: Simultaneously, the state fluctuation amplitude Fj within the current window is calculated, defined as: Fj = maximum value (state code within the window) − minimum value (state code within the window); this difference value Δej is used to determine the direction of state change (positive value indicates a jump to a higher risk level), while the fluctuation amplitude Fj is used to quantify the severity of the state jump.
[0056] Two threshold parameters are set for mutation determination: State transition direction threshold: Δej≥1, indicating a transition from a lower to a higher state level; Fluctuation amplitude threshold: Fj≥T1, where T1 is a preset threshold, and a value of 1 or 2 is recommended. When a certain time point j meets the above two conditions, i.e., the current state level is higher than the previous time period (i.e., Δej≥1); and the state level changes drastically within the current window (i.e., Fj≥T1), then a mutation in the health state trend is considered to exist at that time point, and this time point is marked as a potential health risk node pj.
[0057] After performing the aforementioned window difference scan on the entire state-encoded sequence, a set of location indices satisfying the mutation condition is recorded: a risk node set P = {p1, p2, ..., pm}, where each pj ∈ {1, 2, ..., n}. For each risk node pj, the following information is output: time index position j; the original health status level label for the corresponding time period; the level change value Δej relative to the previous time period; and the state fluctuation amplitude Fj within the current window. This information can serve as the triggering basis for downstream modules to make health intervention decisions.
[0058] By employing sliding window difference analysis and mutation condition determination, this embodiment can dynamically identify potential risk transition points in a continuous health state sequence, capturing key time nodes of mutations from low-risk to high-risk. It boasts advantages such as strong real-time performance, low computational overhead, and high sensitivity. Compared to traditional methods that rely solely on single-point state classification, this method focuses more on state change trends and fluctuation patterns, effectively identifying latent risk increases, chronic disease progression, or sudden health deterioration, providing technical support for timely intervention recommendations.
[0059] The intervention suggestion generation module generates intervention suggestions based on the changing trends of the fusion features corresponding to the health risk node P.
[0060] After identifying the set of health risk nodes, for each risk node, its position index in the original time series is determined. Then, using this index as the center, several time periods are extended forward and backward (e.g., two time periods each). A multi-dimensional fused feature vector is extracted from this interval to form a continuous feature change sequence, represented as: Fusion Feature Change Sequence. ;in: This represents the fused feature vector during the risk node time period; each fused feature vector contains normalized physiological features, behavioral features, and environmental features.
[0061] Trend analysis is performed on the fused feature change sequence. Linear fitting or moving average methods are used to calculate the direction and magnitude of change of each feature within the time window. The specific judgment logic is as follows: 1) Abnormal growth identification: When a feature value shows a continuous upward trend within the current window, and the cumulative growth exceeds the set growth threshold T1, the feature is considered to have abnormal growth; 2) Abnormal decline identification: When a feature value continuously declines and the magnitude exceeds the set decline threshold T2, it is marked as an abnormal decline; 3) Continuous deviation identification: When a feature value remains outside the normal range (e.g., exceeding the upper limit or falling below the lower limit) for multiple consecutive time periods and the time exceeds the set threshold T3 (e.g., 3 time periods), it is judged as continuous deviation.
[0062] The thresholds T1, T2, and T3 mentioned above can be set based on population baseline data or individual historical statistical values, such as a heart rate increase of more than 20%, a blood oxygen decrease of more than 5%, or noise levels exceeding 80 decibels for more than 15 minutes.
[0063] Establish a predefined rule base for matching intervention strategies. Each rule in the rule base includes: the name of the abnormal feature; the type of abnormal trend (increase, decrease, or continuous deviation); the corresponding explanation of the probability of health risk; and the recommended intervention suggestions.
[0064] The abnormal trend detected in the current risk node is compared with the matching items in the rule base, and corresponding intervention suggestions are extracted based on the matching results.
[0065] For example, if "heart rate continues to rise + sleep repair efficiency decreases", the rule "physiological over-activation stress" in the rule base is matched, and the recommended suggestion is "rest and adjustment, reduce stimulating diet, and consider relaxation training"; if "blood oxygen continues to be low + noise continues to be high", the rule "insufficient oxygen supply caused by environmental factors" is matched, and the recommended suggestion is "improve environmental noise and provide intermittent oxygen supply if necessary".
[0066] The matching results will generate standardized intervention recommendation text, which includes: recommendation type: such as "behavior adjustment", "sleep optimization", "medical advice"; recommendation content: such as "appropriately increase lunch break time" or "recommend medical examination of cardiopulmonary function"; trigger description: list the feature name, trend type and degree of abnormality that triggered the intervention recommendation.
[0067] For example, intervention recommendations: It is recommended to reduce high-intensity physical activity and increase rest intervals within the next 12 hours.
[0068] Triggering cause: The heart rate was detected to increase by more than 20% in three consecutive time periods, while the sleep repair efficiency was less than 30% of the personal baseline.
[0069] The final intervention recommendations can be displayed on the user terminal interface or remote health monitoring platform for reference by the user, family members, or health management personnel.
[0070] By employing multidimensional analysis of the changing trends of fusion features and a rule-driven matching method, this embodiment can dynamically generate structured and targeted intervention suggestions based on changes in multi-source data before and after an individual's health risk points, effectively enhancing the proactiveness and practicality of the health monitoring system. This method combines personalization, scalability, and interpretability, making it suitable for continuous health management and intelligent early warning in various scenarios.
[0071] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A health information monitoring and management system based on multi-source data fusion analysis, characterized in that: include: The data acquisition module collects multi-source raw health data of a preset object within a preset time period. The multi-source raw health data includes a first type of physiological signal data, a second type of behavioral characteristic data, and a third type of environmental exposure data. The data preprocessing module performs standardization processing on the first, second and third types of data respectively to obtain the corresponding normalized feature sequences S1, S2 and S3; The feature fusion module constructs a multi-dimensional fusion feature vector set F, where each feature is calculated from data analysis within the corresponding time window of normalized S1, S2, and S3. The health status assessment module, based on the multi-dimensional fusion feature vector set F, inputs the trained health status classification model and outputs a health status level label L corresponding to each time period. The risk detection module, based on continuous health status level labels L, dynamically tracks regions where health status changes abruptly in the time series, identifying potential health risk nodes P={p1,p2,...,pj,...,pm}, where each pj represents a mutation start point or trend inflection point. The intervention suggestion generation module generates intervention suggestions based on the changing trends of the fusion features corresponding to the health risk node P.
2. The health information monitoring and management system based on multi-source data fusion analysis according to claim 1, characterized in that: The first type of physiological signal data includes heart rate, blood oxygen saturation, and respiratory rate data.
3. The health information monitoring and management system based on multi-source data fusion analysis according to claim 1, characterized in that: The second category of behavioral characteristic data includes steps, exercise intensity, sleep structure, and number of times the body turns over.
4. The health information monitoring and management system based on multi-source data fusion analysis according to claim 1, characterized in that: The third category of environmental exposure data includes noise intensity, ambient temperature and humidity, and atmospheric pressure.
5. The health information monitoring and management system based on multi-source data fusion analysis according to claim 1, characterized in that: The data preprocessing module includes data cleaning, missing value interpolation, time alignment, and standardization. The standardization process includes normalizing each feature value to the [0,1] interval.
6. The health information monitoring and management system based on multi-source data fusion analysis according to claim 1, characterized in that: The feature fusion module constructs a tensor structure to perform non-negative tensor decomposition on the normalized feature sequence, extracts latent stress factors, and calculates the composite stress load index for each time period as part of the multidimensional fusion feature vector group F.
7. The health information monitoring and management system based on multi-source data fusion analysis according to claim 6, characterized in that: The feature fusion module further includes constructing a counterfactual sleep scenario based on sleep structure and pre-existing physiological state, combining it with a physiological recovery prediction model, calculating the sleep repair efficiency index for each time period, and using it as part of the multidimensional fusion feature vector group F.
8. The health information monitoring and management system based on multi-source data fusion analysis according to claim 7, characterized in that: The health status assessment module uses a random forest classification model constructed with a supervised learning algorithm. Based on the composite stress load index and sleep repair efficiency index, it outputs the health status level for the corresponding time period. The health status level includes three types: normal, sub-healthy, and abnormal warning.
9. The health information monitoring and management system based on multi-source data fusion analysis according to claim 8, characterized in that: The risk detection module performs sliding window difference analysis on the health status level marker sequence, and identifies the locations where health status levels change abruptly as potential health risk nodes by combining the set thresholds for the transition direction and fluctuation amplitude of the status levels.
10. The health information monitoring and management system based on multi-source data fusion analysis according to claim 9, characterized in that: The intervention suggestion generation module analyzes the trend of changes in the integrated characteristics of the time periods before and after the health risk node, identifies the characteristic patterns of abnormal growth, abnormal decline or continuous deviation from the normal range, and generates intervention suggestions containing suggestion type, suggestion content and triggering reason explanation according to the predefined intervention rule base.