Health monitoring method and device based on Internet of Things

Abstract

The invention relates to the technical field of health monitoring, in particular to a health monitoring method and device based on the Internet of Things, and the method comprises the steps: obtaining a historical health monitoring data set according to an Internet of Things sensor network; dividing the monitoring area into a multi-scale space-time health state grid; and performing state classification through the historical health monitoring data set according to a density peak clustering algorithm to obtain a health state mode. According to the method, the monitoring area is divided into the multi-scale space-time health state grids, and the mapping relation between the environment parameters and the health indexes is combined, so that the microenvironment-health association which is difficult to find by a traditional monitoring system can be captured, for example, in a ward scene, the monitoring accuracy is improved; the phenomenon that the infection risk of a patient is increased due to poor air circulation in a ward can be accurately identified, so that environmental parameters are optimized in a targeted manner; and through collaborative analysis of a global channel and a local channel, a group trend and individual anomaly can be captured.

Description

Technical Field

[0001] The present invention relates to the field of health monitoring technology, and specifically to a health monitoring method and device based on the Internet of Things. Background Art

[0002] Traditional methods usually rely on static environmental parameters or data of a single time dimension and lack the ability to analyze multi-scale spatiotemporal grids, which means that they cannot accurately capture the complex dynamic relationship between environmental factors (such as air quality) and individual health status; and traditional methods often ignore the impact of different times, places and environmental changes on individual health, resulting in the inability to timely detect potential health risks; and traditional methods usually only rely on a single health monitoring indicator, such as heart rate or blood pressure, and do not combine environmental factors (such as temperature and humidity, air quality, etc.) with health data for comprehensive analysis, which makes them unable to accurately identify factors such as increased infection risks caused by poor air quality in wards, or The changes in individual health status under different environmental conditions; moreover, traditional methods usually adopt a unified health monitoring model, which fails to take into account the specific health needs and physiological characteristics of each individual. They often cannot dynamically adjust intervention measures according to changes in individual health patterns, nor do they use real-time physiological data and location information to match personalized health status, thus lacking flexibility and personalization; and traditional methods mostly focus on a single health indicator or local data, and cannot conduct collaborative analysis between the global (such as ward-level rehabilitation status) and the local (such as individual physiological characteristics). Therefore, they cannot fully capture group health trends and individual health anomalies, and cannot achieve more efficient health monitoring and management. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to overcome the shortcomings of the above-mentioned prior art and provide a health monitoring method and device based on the Internet of Things.

[0004] The technical solution for solving the above technical problems is: a health monitoring method based on the Internet of Things, including: Acquire a historical health monitoring dataset based on an Internet of Things sensor network; divide the monitoring area into multi-scale spatiotemporal health status grids, wherein each spatiotemporal health status grid is used to associate a mapping relationship between environmental parameters and health indicators; Performing state classification on the historical health monitoring data set according to a density peak clustering algorithm to obtain a health state pattern, wherein the health state pattern includes a normal pattern set and an abnormal pattern set, and generating a health pattern library based on the health state pattern, wherein each pattern in the health pattern library corresponds to a characterized spatiotemporal anchor point sequence, wherein the spatiotemporal anchor point is a spatiotemporal location node where a significant change in health state occurs; When a mobile monitored object enters the monitoring area, real-time physiological parameters and real-time location information are collected based on the wearable device, and the real-time location information is mapped to the spatiotemporal health status grid; the real-time physiological parameters are feature extracted based on the convolutional neural network to obtain physiological features, and the physiological features are embedded in the spatiotemporal anchor points of the health model library in combination with the transfer learning mechanism; The physiological features are aggregated according to the dual-channel attention mechanism to obtain a health status representation; the matching degree of the health status representation is calculated with the candidate patterns in the health pattern library according to maximum likelihood estimation to obtain a health status pattern sequence that matches the current moment.

[0005] Preferably, the method further comprises: When a sudden change in the physiological parameter is detected, the spatiotemporal anchor point sequence verification process is activated, and the historical spatiotemporal anchor points are retrospectively verified according to the forward-backward algorithm; According to the particle filtering algorithm, combined with the digital twin technology, a virtual health image of the mobile monitoring object is constructed through the health status pattern sequence matching the current moment; The virtual health image is used to preview the evolution paths of different health status patterns in a virtual space to generate multiple possible branches of the individual health trajectory of the mobile monitoring object; When it is detected that the individual's health trajectory deviates from the normal mode threshold, an early warning mechanism is triggered.

[0006] Preferably, the historical health monitoring dataset includes spatiotemporal trajectory data of group objects and corresponding physiological parameter sequences; the dual-channel attention mechanism includes a global channel and a local channel; wherein, the global channel analyzes the spatiotemporal propagation characteristics of the group health situation based on a graph convolutional network, and the local channel tracks the temporal evolution of individual health status based on a long short-term memory network.

[0007] Preferably, the monitoring area is divided into a multi-scale spatiotemporal health status grid, including: Construct a multi-scale spatiotemporal health status grid based on the geographical characteristics of the monitoring area and the health monitoring needs, wherein the multi-scale spatiotemporal health status grid includes a geographic grid with decreasing spatial levels and a time period grid with subdivided time granularity; The spatiotemporal boundary coordinates of each spatiotemporal health state are defined, wherein the spatiotemporal boundary coordinates include a latitude and longitude range, an altitude interval, and a timestamp interval; a cluster analysis is performed on the historical health monitoring dataset based on the K-means algorithm to determine the optimal division granularity of the spatiotemporal health state grid at each scale, so that the correlation strength between the environmental parameters and health indicators within the spatiotemporal health state grid meets a preset threshold.

[0008] Preferably, the monitoring area is divided into a multi-scale spatio-temporal health status grid, which also includes: Deploying Internet of Things sensing nodes inside each of the spatio-temporal health status grids to collect environmental parameters in real time, wherein the environmental parameters include air quality index, temperature and humidity, ultraviolet intensity, and noise decibel value; Synchronously acquiring health indicators of monitoring objects within the spatio-temporal health status grids, wherein the health indicators include blood pressure, heart rate variability, and sleep quality score; Building an environment-health coupling analysis model, wherein the environment-health coupling analysis model adopts a BP neural network to identify the coupling strength between the environmental parameters and the health indicators according to the BP neural network, and to generate a mapping relationship matrix of each of the spatio-temporal health status grids according to the coupling strength.

[0009] Preferably, a health pattern library is generated according to the health status patterns, including: Reconstructing the spatio-temporal trajectories of each pattern in the health pattern library, and aligning the spatio-temporal trajectories of different mobile monitoring objects within the same pattern using a dynamic time warping algorithm; Identifying significant change points in the spatio-temporal trajectories by a turning point detection algorithm, wherein the significant change points are defined as spatio-temporal positions where the absolute value of the derivative of the physiological parameter exceeds 3 times the standard deviation of the mean value within the same pattern and the duration exceeds a preset time; Marking the detected significant change points as spatio-temporal anchor points, and arranging the anchor points within the same pattern in chronological order to obtain a characteristic spatio-temporal anchor point sequence, wherein the spatio-temporal anchor point sequence includes anchor point coordinates, occurrence time stamp, and associated physiological parameter mutation value.

[0010] Preferably, when the physiological parameter mutation is detected, a spatio-temporal anchor point sequence verification process is activated to perform backtracking verification on historical spatio-temporal anchor points, including: In response to a mutation trigger signal, extracting a spatio-temporal anchor point sequence associated with the current spatio-temporal position from the health pattern library; Screening candidate spatio-temporal anchor points with a distance less than a preset distance threshold from the current spatio-temporal position, and arranging the candidate spatio-temporal anchor points in reverse chronological order to form a to-be-verified anchor point sequence; Building a hidden Markov model of the spatio-temporal anchor points, wherein the state space of the hidden Markov model includes a normal pattern set and an abnormal pattern set in the health pattern library, and the physiological feature vector of the to-be-verified spatio-temporal anchor point sequence is observed.

[0011] Preferably, when the physiological parameter mutation is detected, a spatio-temporal anchor point sequence verification process is activated to perform backtracking verification on historical spatio-temporal anchor points, further including: Calculating the probability of being in each state at each time to generate a forward probability matrix, and calculating the probability of observation under a given state to generate a backward probability matrix; Calculating the state posterior probability of each spatiotemporal anchor point to be verified based on the forward probability matrix and the backward probability matrix, and decoding the optimal state path using the Viterbi algorithm; When a preset number of abnormal mode states appear continuously in the optimal state path, backtracking verification is started, wherein the backtracking verification includes a two-level verification mechanism, wherein the first-level verification performs localized anchor point rematching through the edge computing node, and the second-level verification calls the spatiotemporal anchor points of adjacent areas for collaborative verification.

[0012] Preferably, according to the particle filtering algorithm, combined with the digital twin technology, a virtual health image of the mobile monitoring object is constructed through the health status pattern sequence matching the current moment, including: Constructing a high-dimensional health state space in the virtual health image, wherein the high-dimensional health state space is divided into a normal state subspace and multiple abnormal state subspaces based on critical thresholds of physiological parameters; Mapping the current state vector of the virtual healthy image to an initial particle set in a healthy state space; Embedding a state transition prediction model in the virtual health image, wherein the embedded state transition prediction model adopts a long short-term memory network; the state transition prediction model is used to learn the state transition law in the historical health state pattern sequence; Taking the current particle set as input, a predicted particle set is generated through a state transition prediction model, where the predicted particle set includes a state evolution path of multiple time steps.

[0013] The technical solution for solving the above technical problems is: a health monitoring device based on the Internet of Things, which is applicable to the health monitoring method based on the Internet of Things, comprising: A region mapping unit is configured to obtain a historical health monitoring dataset based on an Internet of Things sensor network; divide the monitoring area into multi-scale spatiotemporal health status grids, wherein each spatiotemporal health status grid is configured to associate a mapping relationship between an environmental parameter and a health indicator; a state classification unit, the state classification unit being configured to perform state classification on the historical health monitoring data set according to a density peak clustering algorithm to obtain a health state pattern, wherein the health state pattern includes a normal pattern set and an abnormal pattern set, and to generate a health pattern library according to the health state pattern, wherein each pattern in the health pattern library corresponds to a characterized spatiotemporal anchor point sequence, wherein the spatiotemporal anchor point is a spatiotemporal location node where a significant change in health state occurs; A feature embedding unit, configured to collect real-time physiological parameters and real-time location information of a mobile monitored object according to a wearable device when the mobile monitored object enters the monitoring area, map the real-time location information to the spatiotemporal health status grid, extract features of the real-time physiological parameters according to a convolutional neural network to obtain physiological features, and embed the physiological features into the spatiotemporal anchor points of the health pattern library in combination with a transfer learning mechanism; a health matching unit, configured to aggregate the physiological features according to a dual-channel attention mechanism to obtain a health status representation; and calculate a matching degree between the health status representation and candidate patterns in the health pattern library according to maximum likelihood estimation to obtain a health status pattern sequence that matches the current moment; A retrospective verification unit, configured to activate a spatiotemporal anchor point sequence verification process when a sudden change in the physiological parameter is detected, and perform retrospective verification on the historical spatiotemporal anchor points according to a forward-backward algorithm; A virtual modeling unit, configured to construct a virtual health image of the mobile monitoring object using the health status pattern sequence matching the current moment according to a particle filtering algorithm in combination with digital twin technology; The virtual health image is used to preview the evolution paths of different health status patterns in a virtual space to generate multiple possible branches of the individual health trajectory of the mobile monitoring object; A health warning unit is configured to trigger a warning mechanism when it is detected that the individual's health trajectory deviates from a normal mode threshold.

[0014] The beneficial effects of the present application are as follows: (1) The present application can capture the micro environment-health correlation that the traditional monitoring system cannot find by dividing the monitoring area into a multi-scale space-time health grid and combining the mapping relationship between environmental parameters (such as temperature and humidity, air quality) and health indicators (such as heart rate, blood pressure). For example, in a ward scene, the phenomenon that the infection risk of patients in a ward increases due to poor air circulation can be accurately identified, so as to optimize the environmental parameters; and through the cooperative analysis of the global channel (ward-level rehabilitation situation) and the local channel (individual physiological characteristics), the method can capture both group trends (such as the increase of infection risk in a ward) and individual abnormalities (such as the sudden drop of blood oxygen of a patient), for example, in an ICU scene, it can be distinguished in real time whether the blood oxygen of a patient decreases due to environmental factors (such as poor air quality in the ward) or individual factors (such as lung complications); and when a physiological parameter mutation is detected, the starting point of the change of health status (such as the appearance of the pre-symptom of infection of a patient on the 3rd day after surgery at 14:20) can be accurately located through the backtracking verification of the historical space-time anchor point by the forward-backward algorithm; (2) The present application provides a multi-dimensional analysis framework for health status monitoring at different locations and time periods by dividing the monitoring area into a multi-scale space-time health grid and combining the mapping relationship between environmental parameters and health indicators, which can more flexibly and comprehensively evaluate the health risk; and by combining the real-time physiological parameters collected by the wearable device with the real-time location information, the health status of the mobile monitoring object can be dynamically monitored, and according to the physiological characteristics extracted by the convolutional neural network, the health status can be accurately matched by combining the health pattern library, which helps to provide personalized health management for each individual; (3) The present application can predict the evolution path of the individual health status by using the particle filtering algorithm and the digital twin technology to construct a virtual health mirror, which can identify potential risks in advance and provide preventive health intervention solutions before the actual occurrence of health abnormalities, and when the individual health trajectory deviates from the normal mode threshold, the early warning mechanism can be triggered quickly to provide immediate response for health management personnel, thereby reducing the occurrence of health risks. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A step flow diagram of the overall method in an embodiment proposed by the present application is shown in the figure; Figure 2 A device architecture diagram of the overall device in an embodiment proposed by the present application is shown in the figure.

[0016] Reference signs: 1, area mapping unit; 2, state classification unit; 3, feature embedding unit; 4, health matching unit; 5, backtracking verification unit; 6, virtual modeling unit; 7, health warning unit. DETAILED DESCRIPTION

[0017] Embodiment one, as Figure 1As shown, the present invention proposes a health monitoring method based on the Internet of Things, including: S1. Obtain a historical health monitoring dataset based on an IoT sensor network; divide the monitoring area into multi-scale spatiotemporal health status grids, where each spatiotemporal health status grid is used to associate a mapping relationship between environmental parameters and health indicators; S2. Classify the historical health monitoring data set using a density peak clustering algorithm to obtain health status patterns, wherein the health status patterns include a normal pattern set and an abnormal pattern set. Generate a health pattern library based on the health status patterns, wherein each pattern in the health pattern library corresponds to a characterized spatiotemporal anchor point sequence, wherein the spatiotemporal anchor point is a spatiotemporal location node where the health status undergoes a significant change; S3. When a mobile monitored subject enters the monitoring area, real-time physiological parameters and real-time location information are collected using wearable devices, and the real-time location information is mapped to the spatiotemporal health status grid. A convolutional neural network is used to extract features from the real-time physiological parameters to obtain physiological features, which are then embedded into the spatiotemporal anchor points of the health model library using a transfer learning mechanism. S4. Aggregate physiological features according to the dual-channel attention mechanism to obtain a health status representation; calculate the matching degree between the health status representation and the candidate patterns in the health pattern library according to maximum likelihood estimation to obtain a health status pattern sequence that matches the current moment.

[0018] In this invention, data is collected collaboratively through an IoT sensor network (such as ambient temperature and humidity sensors, air quality monitors, and ultraviolet disinfection equipment status sensors) and wearable devices (such as smart bracelets and electrocardiogram patches) to achieve spatiotemporal alignment of environmental parameters and physiological indicators. Data types include: environmental parameters (temperature and humidity, air quality index, and disinfection equipment operation time), physiological indicators (heart rate variability, blood pressure rhythm, and sleep apnea episodes), and location coordinates (UWB positioning tags with an accuracy of 0.3 meters). Spatial stratification uses a three-level grid: the first grid covers the entire ward floor (50m×30m), the second grid corresponds to a single ward (8m×6m), and the third grid focuses on a 1m³ area around the bed, achieving dynamic adaptation of spatial granularity. Time granularity is set to 60 seconds, 10 minutes, and 1 hour to meet the monitoring needs of different physiological processes (e.g., sudden arrhythmia requires second-level monitoring, while chronic disease trend analysis requires hour-level data). Environment-health mapping: Establish association rules between environmental parameters and health indicators, for example, when the AQI of a ward is greater than 150 and the humidity is greater than 75%; Global channel: Analyze ward-level rehabilitation status through a graph attention network (GAT), with nodes representing the current risk level of each ward and edge weights representing the frequency of patient movement (for example, when a patient is transferred from a high-risk ward, the edge weight increases by 0.3); Local channel: Use a Transformer encoder to process individual patient characteristics, and a multi-head attention mechanism to focus on segments where SpO2 decreases (the query vector is generated by a clinical threshold, such as a high weight is triggered when SpO2 < 90%); Fusion output: Generate a 128-dimensional health status representation vector, of which the first 64 dimensions are group status codes and the last 64 dimensions are individual abnormality indices.

[0019] In an optional embodiment, the method further includes: S5. When a sudden change in physiological parameters is detected, the spatiotemporal anchor point sequence verification process is activated, and the historical spatiotemporal anchor points are back-tested according to the forward-backward algorithm; S6. Based on the particle filter algorithm and combined with digital twin technology, a virtual health image of the mobile monitoring object is constructed through a health status pattern sequence that matches the current moment; Among them, the virtual health mirror is used to preview the evolution paths of different health status patterns in the virtual space to generate multiple possible branches of the individual health trajectory of the mobile monitoring object; S7. When it is detected that the individual health trajectory deviates from the normal mode threshold, the early warning mechanism is triggered.

[0020] Example 2. A health monitoring method based on the Internet of Things proposed by the present invention, compared with Example 1, this embodiment also includes: a historical health monitoring data set includes spatiotemporal trajectory data of group objects and corresponding physiological parameter sequences; a dual-channel attention mechanism includes a global channel and a local channel; wherein, the global channel analyzes the spatiotemporal propagation characteristics of the group health situation based on the graph convolutional network, and the local channel tracks the temporal evolution of the individual health status based on the long short-term memory network.

[0021] In an optional embodiment, the monitoring area is divided into a multi-scale spatiotemporal health status grid, including: A1. Construct a multi-scale spatiotemporal health status grid based on the geographic characteristics of the monitoring area and health monitoring needs. The multi-scale spatiotemporal health status grid includes a geographic grid with decreasing spatial levels and a time period grid with finer temporal granularity. A2. Define the spatiotemporal boundary coordinates for each spatiotemporal health state, where the spatiotemporal boundary coordinates include latitude and longitude ranges, altitude intervals, and timestamp intervals. Perform cluster analysis on historical health monitoring datasets based on the K-means algorithm to determine the optimal granularity of spatiotemporal health state grids at each scale, so that the correlation strength between environmental parameters and health indicators within the spatiotemporal health state grids meets the preset threshold.

[0022] It should be noted that in health monitoring, there is a close relationship between environmental parameters (such as temperature, humidity, air quality, etc.) and health indicators (such as heart rate, blood pressure, respiratory rate, etc.). Through the K-means algorithm, the characteristics of different health states can be determined based on historical data, and then the optimal granularity of each grid can be determined; the choice of division granularity needs to be determined based on the correlation strength of the data, that is, whether the relationship strength between environmental factors and health indicators in a certain grid meets the preset threshold. If the relationship is not strong enough, it may be necessary to adjust the grid granularity or increase the accuracy of the data to obtain better monitoring results.

[0023] In an optional embodiment, the monitoring area is divided into a multi-scale spatiotemporal health status grid, further comprising: A3. Deploy IoT sensor nodes within each spatiotemporal health status grid to collect environmental parameters in real time, including air quality index, temperature and humidity, UV intensity, and noise decibel level. A4. Synchronously obtain health indicators of the monitored subjects within the spatiotemporal health status grid, including blood pressure, heart rate variability, and sleep quality score; A5. Construct an environment-health coupling analysis model. The environment-health coupling analysis model uses a BP neural network to identify the coupling strength between environmental parameters and health indicators. Based on the coupling strength, a mapping relationship matrix for each spatiotemporal health status grid is generated.

[0024] It should be noted that the BP (back propagation) neural network is a commonly used deep learning algorithm, which is suitable for complex data pattern recognition and prediction, and in this technical feature, the BP neural network is used to analyze the coupling relationship between environmental parameters and health indicators; the identification of coupling strength can identify the influence of different environmental parameters (such as temperature and humidity, air quality, noise, etc.) on health indicators (such as blood pressure, heart rate, sleep quality, etc.). Coupling strength is a key indicator for measuring the impact of environmental changes on health status; after training the BP neural network and obtaining the coupling strength, the generated mapping relationship matrix can clearly represent the correlation between environmental parameters and health indicators in each spatio-temporal health status grid, which provides a basis for the development of health intervention measures and environmental regulation.

[0025] In an optional embodiment, generating a health pattern library according to the health status pattern includes: B1, reconstructing the spatio-temporal trajectory of each pattern in the health pattern library, and aligning the spatio-temporal trajectories of different mobile monitoring objects in the same pattern using a dynamic time warping algorithm; B2, identifying significant change points in the spatio-temporal trajectory by a turning point detection algorithm, wherein the significant change point is defined as a spatio-temporal position with a physiological parameter derivative absolute value exceeding 3 times the standard deviation of the mean value in the same pattern and a duration exceeding a preset time; B3, marking the detected significant change points as spatio-temporal anchor points, and arranging the anchor points in the same pattern in chronological order to obtain a characteristic spatio-temporal anchor point sequence, wherein the spatio-temporal anchor point sequence includes anchor point coordinates, time stamp and associated physiological parameter mutation value.

[0026] It should be noted that dynamic time warping (DTW) is an algorithm commonly used for time series alignment, especially suitable for processing nonlinearly changing time series data. In health monitoring, different monitoring objects (such as different people's health indicators) may exhibit different trends at different times, but these changes essentially belong to the same pattern. The DTW algorithm aligns these spatio-temporal trajectories so that health data of different monitoring objects under the same pattern can be compared and analyzed; the turning point detection algorithm is used to identify mutations or change points in the spatio-temporal trajectory, especially those representing significant changes in health status. In health monitoring, turning points usually mean changes in physiological parameters such as heart rate, blood pressure, body temperature, etc., which may suddenly become abnormal or change, which is crucial for health intervention and early warning; by calculating the derivative absolute value of the physiological parameter, the speed and amplitude of the parameter change can be evaluated. When the speed (i.e. derivative) of a physiological parameter change abnormally increases, it means that there may be a sudden event or abnormality in health. Such change points are judged by comparing them with the average value and standard deviation in the pattern.

[0027] In an optional embodiment, when a physiological parameter mutation is detected, a spatio-temporal anchor point sequence verification process is activated, and a historical spatio-temporal anchor point is verified in a backtracking manner, including: C1, in response to a mutation trigger signal, extracting a spatio-temporal anchor point sequence associated with a current spatio-temporal position from a health mode library; C2, screening out candidate spatio-temporal anchor points with a distance less than a preset distance threshold from the current spatio-temporal position, and arranging them in a time reverse order to form a to-be-verified anchor point sequence; C3, constructing a hidden Markov model of the spatio-temporal anchor point, wherein a state space of the hidden Markov model includes a normal mode set and an abnormal mode set in the health mode library, and a physiological feature vector of the to-be-verified spatio-temporal anchor point sequence is observed.

[0028] It should be noted that the mutation trigger signal refers to a signal generated when a physiological parameter (such as heart rate, blood pressure, body temperature, etc.) changes significantly; the hidden Markov model is a statistical model widely used in time series data analysis, especially suitable for inferring hidden states in sequence data, in health monitoring, HMM can be used to infer the hidden state (such as normal or abnormal state) of the health state from the observation data; the current state is inferred by analyzing the observed physiological feature vectors (such as heart rate, blood pressure changes, etc.) in the to-be-verified spatio-temporal anchor point sequence, and these feature vectors are used as observation data in HMM to infer the hidden state that the system may be in.

[0029] In an optional embodiment, when a physiological parameter mutation is detected, a spatio-temporal anchor point sequence verification process is activated, and a historical spatio-temporal anchor point is verified in a backtracking manner, including: C4, calculating the probability of being in each state at each time, generating a forward probability matrix, and calculating the probability of observation under a given state, generating a backward probability matrix; C5, calculating the state posterior probability of each to-be-verified spatio-temporal anchor point based on the forward probability matrix and the backward probability matrix, and decoding the optimal state path using the Viterbi algorithm; C6, when a preset number of abnormal mode states appear continuously in the optimal state path, starting backtracking verification, wherein the backtracking verification includes a two-level verification mechanism, wherein the first-level verification performs localized anchor point re-matching through an edge computing node, and the second-level verification calls spatio-temporal anchor points of adjacent areas for collaborative verification.

[0030] It should be noted that the Viterbi algorithm can quickly infer the most likely health state path according to the current observation data, helping the system to accurately predict the health state and diagnose the abnormality; by localizing the anchor point re-matching in the edge computing node, the system can quickly re-evaluate and confirm the state of a certain space-time anchor point, which improves the timeliness of verification and reduces the dependence on centralized computing resources; when the first-level verification cannot confirm the abnormal state, the second-level verification will cooperate with the adjacent area of the space-time anchor point to verify, which means that the system will expand to the adjacent area data for comparison to increase the accuracy of verification. In this way, it can reduce errors caused by regional or single data source problems and improve the reliability of overall anomaly detection.

[0031] In an optional embodiment, according to the particle filtering algorithm, a virtual health mirror of the mobile monitoring object is constructed by combining the digital twin technology with the health state mode sequence matched with the current time, including: D1, constructing a high-dimensional health state space in the virtual health mirror, the high-dimensional health state space is divided into normal state subspaces and multiple abnormal state subspaces with physiological parameter critical threshold as boundary; D2, mapping the current state vector of the virtual health mirror into an initial particle set in the health state space; D3, embedding a state transition prediction model in the virtual health mirror, wherein the embedded state transition prediction model adopts a long short-term memory network; the state transition prediction model is used to learn the state transition law in the historical health state mode sequence; D4, taking the current particle set as input, generating a predicted particle set through the state transition prediction model, the predicted particle set includes state evolution paths of multiple time steps.

[0032] It should be noted that the health state space is divided by the critical threshold of physiological parameters (such as heart rate, blood pressure, body temperature, etc.), which distinguishes normal state from abnormal state. Through the construction of this high-dimensional state space, the health status of the individual can be comprehensively evaluated in multiple dimensions; mapping the current state vector of the virtual health mirror into the health state space can generate an initial particle set, each particle represents a state point in the health state space, and the distribution of the particle set reflects the diversity and possibility of the current health state; by collecting the health data of the individual at different time points, LSTM can learn the conversion law between different physiological states (for example, the trend of heart rate change), thereby making accurate prediction of future state; based on the current particle set and the state transition prediction model, the system can generate a new particle set representing the health state evolution path within multiple time steps, which describes the trend of individual health state change over time.

[0033] Embodiment three, as Figure 2 As shown, the present application proposes a health monitoring device based on Internet of Things, which is suitable for the health monitoring method based on Internet of Things, comprising: The region mapping unit 1 is used to obtain a historical health monitoring data set according to an Internet of Things sensor network; and the monitoring region is divided into a multi-scale space-time health state grid, wherein each space-time health state grid is used to associate the mapping relationship between environmental parameters and health indicators; The state classification unit 2 is used to perform state classification through the historical health monitoring data set according to a density peak clustering algorithm to obtain a health state mode, wherein the health state mode comprises a normal mode set and an abnormal mode set, and a health mode library is generated according to the health state mode, wherein each mode in the health mode library corresponds to a characteristic space-time anchor point sequence, wherein the space-time anchor point is a space-time position node where the health state changes significantly; The feature embedding unit 3 is used to map the real-time location information to the space-time health state grid according to the real-time physiological parameters and real-time location information collected by the wearable device when the mobile monitoring object enters the monitoring region; the physiological features are obtained by performing feature extraction on the real-time physiological parameters according to a convolutional neural network; and the physiological features are embedded into the space-time anchor points of the health mode library in combination with a transfer learning mechanism; The health matching unit 4 is used to aggregate the physiological features to obtain a health state representation according to a double-channel attention mechanism; and the health state representation is matched with a candidate mode in the health mode library to calculate the matching degree according to maximum likelihood estimation, so as to obtain a health state mode sequence matched with the current time; The backtracking verification unit 5 is used to activate the space-time anchor point sequence verification process when the physiological parameter mutation is detected, and perform backtracking verification on the historical space-time anchor points according to a forward-backward algorithm; The virtual modeling unit 6 is used to construct a virtual health mirror of the mobile monitoring object according to a particle filtering algorithm in combination with a digital twin technology through the health state mode sequence matched with the current time; The virtual health mirror is used to pre-act the evolution path of different health state modes in a virtual space to generate multiple possible branches of the individual health trajectory of the mobile monitoring object; The health warning unit 7 is used to trigger a warning mechanism when the individual health trajectory deviates from the normal mode threshold.

[0034] The embodiments of the present application are described in detail above in combination with the drawings, but the present application is not limited thereto, and various changes can be made within the scope of knowledge possessed by those skilled in the art without departing from the purpose of the present application.

Claims

1. A health monitoring method based on the Internet of Things, characterized in that: include: Acquire a historical health monitoring dataset based on an Internet of Things sensor network; divide the monitoring area into multi-scale spatiotemporal health status grids, wherein each spatiotemporal health status grid is used to associate a mapping relationship between environmental parameters and health indicators; Performing state classification on the historical health monitoring data set according to a density peak clustering algorithm to obtain a health state pattern, wherein the health state pattern includes a normal pattern set and an abnormal pattern set, and generating a health pattern library based on the health state pattern, wherein each pattern in the health pattern library corresponds to a characterized spatiotemporal anchor point sequence, wherein the spatiotemporal anchor point is a spatiotemporal location node where a significant change in health state occurs; When a mobile monitored object enters the monitoring area, real-time physiological parameters and real-time location information are collected based on the wearable device, and the real-time location information is mapped to the spatiotemporal health status grid; the real-time physiological parameters are feature extracted based on the convolutional neural network to obtain physiological features, and the physiological features are embedded in the spatiotemporal anchor points of the health model library in combination with the transfer learning mechanism; The physiological features are aggregated according to the dual-channel attention mechanism to obtain a health status representation; the matching degree of the health status representation is calculated with the candidate patterns in the health pattern library according to maximum likelihood estimation to obtain a health status pattern sequence that matches the current moment.

2. The health monitoring method based on the Internet of Things according to claim 1, characterized in that: The method further comprises: When a sudden change in the physiological parameter is detected, the spatiotemporal anchor point sequence verification process is activated, and the historical spatiotemporal anchor points are retrospectively verified according to the forward-backward algorithm; According to the particle filtering algorithm, combined with the digital twin technology, a virtual health image of the mobile monitoring object is constructed through the health status pattern sequence matching the current moment; The virtual health image is used to preview the evolution paths of different health status patterns in a virtual space to generate multiple possible branches of the individual health trajectory of the mobile monitoring object; When it is detected that the individual's health trajectory deviates from the normal mode threshold, an early warning mechanism is triggered.

3. The health monitoring method based on the Internet of Things according to claim 2, characterized in that: The historical health monitoring dataset includes spatiotemporal trajectory data of group objects and corresponding physiological parameter sequences; the dual-channel attention mechanism includes a global channel and a local channel; wherein, the global channel analyzes the spatiotemporal propagation characteristics of the group health status based on a graph convolutional network, and the local channel tracks the temporal evolution of individual health status based on a long-short-term memory network.

4. The health monitoring method based on the Internet of Things according to claim 3, characterized in that: The monitoring area is divided into multi-scale spatiotemporal health status grids, including: Construct a multi-scale spatiotemporal health status grid based on the geographical characteristics of the monitoring area and the health monitoring needs, wherein the multi-scale spatiotemporal health status grid includes a geographic grid with decreasing spatial levels and a time period grid with subdivided time granularity; The spatiotemporal boundary coordinates of each spatiotemporal health state are defined, wherein the spatiotemporal boundary coordinates include a latitude and longitude range, an altitude interval, and a timestamp interval; a cluster analysis is performed on the historical health monitoring dataset based on the K-means algorithm to determine the optimal division granularity of the spatiotemporal health state grid at each scale, so that the correlation strength between the environmental parameters and health indicators within the spatiotemporal health state grid meets a preset threshold.

5. The health monitoring method based on the Internet of Things according to claim 4, characterized in that: The monitoring area is divided into a multi-scale spatiotemporal health status grid, which also includes: Deploy an IoT sensor node in each of the spatiotemporal health status grids to collect environmental parameters in real time, wherein the environmental parameters include air quality index, temperature and humidity, ultraviolet intensity, and noise decibel value; Synchronously acquiring health indicators of the monitored subjects within the spatiotemporal health status grid, the health indicators including blood pressure, heart rate variability, and sleep quality score; An environment-health coupling analysis model is constructed, wherein the environment-health coupling analysis model adopts a BP neural network, identifies the coupling strength between the environmental parameters and the health indicators based on the BP neural network, and generates a mapping relationship matrix for each of the spatiotemporal health status grids based on the coupling strength.

6. The health monitoring method based on the Internet of Things according to claim 5, characterized in that: Generating a health mode library according to the health status mode includes: Reconstructing the spatiotemporal trajectory of each pattern in the health pattern library, and aligning the spatiotemporal trajectories of different mobile monitoring objects in the same pattern using a dynamic time warping algorithm; Identify significant change points in the spatiotemporal trajectory using an inflection point detection algorithm, where the significant change point is defined as a spatiotemporal location where the absolute value of the derivative of the physiological parameter exceeds three standard deviations of the mean within the same pattern and persists for more than a preset time; The detected significant change points are marked as spatiotemporal anchor points, and the anchor points in the same pattern are arranged in chronological order to obtain a characterized spatiotemporal anchor point sequence, which includes anchor point coordinates, occurrence timestamps, and associated physiological parameter mutation values.

7. The health monitoring method based on the Internet of Things according to claim 6, characterized in that: When a sudden change in the physiological parameter is detected, the spatiotemporal anchor point sequence verification process is activated to perform retrospective verification on the historical spatiotemporal anchor points, including: In response to a mutation trigger signal, extracting a spatiotemporal anchor point sequence associated with a current spatiotemporal position from the health pattern library; Filter out candidate spatiotemporal anchor points whose distance from the current spatiotemporal position is less than a preset distance threshold, and arrange them in reverse chronological order to form a sequence of anchor points to be verified; A hidden Markov model of spatiotemporal anchor points is constructed, wherein the state space of the hidden Markov model includes a normal pattern set and an abnormal pattern set in a healthy pattern library, and the physiological feature vector of the spatiotemporal anchor point sequence to be verified is observed.

8. The health monitoring method based on the Internet of Things according to claim 7, characterized in that: When a sudden change in the physiological parameter is detected, the spatiotemporal anchor point sequence verification process is activated to perform retrospective verification on the historical spatiotemporal anchor points, further comprising: Calculate the probability of being in each state at each moment to generate a forward probability matrix, and calculate the probability of observation in a given state to generate a backward probability matrix; Calculating the state posterior probability of each spatiotemporal anchor point to be verified based on the forward probability matrix and the backward probability matrix, and decoding the optimal state path using the Viterbi algorithm; When a preset number of abnormal mode states appear continuously in the optimal state path, backtracking verification is started, wherein the backtracking verification includes a two-level verification mechanism, wherein the first-level verification performs localized anchor point rematching through edge computing nodes, and the second-level verification calls the spatiotemporal anchor points of adjacent areas for collaborative verification.

9. The health monitoring method based on the Internet of Things according to claim 8, characterized in that: According to the particle filter algorithm, combined with the digital twin technology, a virtual health image of the mobile monitoring object is constructed through the health status pattern sequence matching the current moment, including: Constructing a high-dimensional health state space in the virtual health image, wherein the high-dimensional health state space is divided into a normal state subspace and multiple abnormal state subspaces based on critical thresholds of physiological parameters; Mapping the current state vector of the virtual healthy image to an initial particle set in a healthy state space; Embedding a state transition prediction model in the virtual health image, wherein the embedded state transition prediction model adopts a long short-term memory network; the state transition prediction model is used to learn the state transition law in the historical health state pattern sequence; Taking the current particle set as input, a predicted particle set is generated through a state transition prediction model, where the predicted particle set includes a state evolution path of multiple time steps.

10. A health monitoring device based on the Internet of Things, applicable to the health monitoring method based on the Internet of Things according to any one of claims 1 to 9, characterized in that: include: A region mapping unit (1) is used to obtain a historical health monitoring data set based on an Internet of Things sensor network; divide the monitoring area into multi-scale spatiotemporal health status grids, wherein each of the spatiotemporal health status grids is used to associate a mapping relationship between an environmental parameter and a health indicator; A state classification unit (2), the state classification unit (2) is used to perform state classification on the historical health monitoring data set according to a density peak clustering algorithm to obtain a health state pattern, wherein the health state pattern includes a normal pattern set and an abnormal pattern set, and a health pattern library is generated according to the health state pattern, wherein each pattern in the health pattern library corresponds to a characterized spatiotemporal anchor point sequence, wherein the spatiotemporal anchor point is a spatiotemporal position node where a significant change in health state occurs; A feature embedding unit (3), wherein the feature embedding unit (3) is used to collect real-time physiological parameters and real-time location information according to the wearable device when the mobile monitoring object enters the monitoring area, and map the real-time location information to the spatiotemporal health status grid; extract features of the real-time physiological parameters according to a convolutional neural network to obtain physiological features, and embed the physiological features into the spatiotemporal anchor points of the health model library in combination with a transfer learning mechanism; A health matching unit (4), the health matching unit (4) is used to aggregate the physiological features according to a dual-channel attention mechanism to obtain a health state representation; and calculate the matching degree between the health state representation and the candidate patterns in the health pattern library according to maximum likelihood estimation to obtain a health state pattern sequence that matches the current moment; A retrospective verification unit (5), wherein the retrospective verification unit (5) is used to activate the spatiotemporal anchor point sequence verification process when a sudden change in the physiological parameter is detected, and to perform retrospective verification on the historical spatiotemporal anchor points according to a forward-backward algorithm; A virtual modeling unit (6), the virtual modeling unit (6) is used to construct a virtual health image of the mobile monitoring object through the health status pattern sequence matching the current moment according to a particle filtering algorithm combined with digital twin technology; The virtual health image is used to preview the evolution paths of different health status patterns in a virtual space to generate multiple possible branches of the individual health trajectory of the mobile monitoring object; A health warning unit (7) is used to trigger a warning mechanism when it is detected that the individual's health trajectory deviates from a normal mode threshold.

Images