Intelligent large health system and data processing method thereof

Abstract

The invention relates to the technical field of data processing and analysis, in particular to an intelligent large health system and a data processing method thereof, and the method comprises the steps: collecting the data of a multi-source heterogeneous data source in real time, and comprehensively capturing the multi-dimensional health data of a user; a multi-modal health knowledge graph is dynamically constructed and optimized through an efficient data preprocessing and feature alignment technology, and the integration value and the utilization efficiency of data are improved; a graph neural network and a time sequence analysis model are utilized to realize accurate evaluation of the health state of the user and timely prediction of future risks, and predictability and pertinence of health management are effectively improved; based on a reinforcement learning algorithm, a highly personalized intervention sequence can be generated according to the individual condition of a user, the accuracy of health intervention is enhanced, and the positive change of the health behavior of the user is promoted; an intervention instruction is executed through intelligent equipment, user feedback is monitored in real time, an intervention strategy is dynamically adjusted, and the flexibility and adaptability of intervention measures are ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing analysis, and in particular to a smart health system and a data processing method thereof. BACKGROUND

[0002] In the current health management field, the following problems still exist: under the traditional health management mode, the multi-source data of user's physiological parameters, behavior activities, environmental perception and medical records are often independently stored by different systems or devices, forming data islands, and it is difficult to realize the integration and comprehensive utilization of data; traditional health assessment methods mostly rely on manual inspection and user self-reporting, which are easily affected by subjective factors, and it is difficult to fully capture the real-time health status of users, and there is subjectivity and lag in the assessment, which may lead to deviation in the health assessment results; traditional health risk prediction models are usually based on limited historical data and static indicators, and it is difficult to dynamically reflect the real-time changes of user's health status, and it is difficult to predict potential health risks in time; traditional health intervention strategies often use general schemes, ignoring the individual differences between users and the dynamic changes of health status, and it is difficult to effectively meet the specific needs of users, affecting the intervention effect, and limiting the continuous optimization and improvement of health management services. Therefore, the present application proposes a smart health system and a data processing method thereof. SUMMARY

[0003] The purpose of the present application is to solve the problems in the background art and propose a smart health system and a data processing method thereof.

[0004] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions: A smart health system, comprising: a multi-source data acquisition module, a data fusion and knowledge graph construction module, a health status assessment and risk prediction module, a personalized intervention strategy generation module, and an intervention execution and feedback adaptation module; The multi-source data acquisition module is used to acquire the physiological parameter data, behavior activity data, environmental perception data and medical record data of the user in real time or quasi-real time from multiple heterogeneous data sources; The data fusion and knowledge graph construction module is used to preprocess, extract and align the collected multi-source data, and construct and dynamically update a multi-modal health knowledge graph based thereon, wherein the nodes represent entities including users, physiological indicators, symptoms, behaviors, environmental factors and diseases, and the edges represent the semantic relationship and statistical relationship strength between entities; a health state assessment and risk prediction module configured to learn embedding representations of nodes and relationships based on the multi-modal health knowledge graph using a graph neural network model, and generate real-time health state assessment results and future short-term health risk prediction results of an individual user in combination with a time series analysis model; a personalized intervention strategy generation module configured to infer and generate a personalized intervention sequence for a target user including a series of ordered intervention actions based on the health state assessment results and the health risk prediction results by using a dynamic strategy optimization algorithm based on reinforcement learning; an intervention execution and feedback adaptation module configured to convert the personalized intervention sequence into execution instructions matching the interface protocol of the smart device, drive the connected smart device to execute the intervention, and monitor the response data of the target user to the intervention in real time, and dynamically adjust the intervention strategy and the relationship strength of the multi-modal health knowledge graph according to the response data.

[0005] Further, the data fusion and knowledge graph construction module pre-processes, extracts features, and aligns the collected multi-source data, and the process of constructing and dynamically updating the multi-modal health knowledge graph includes: pre-processing the multi-source data, including data cleaning, denoising, outlier detection and filling, and format standardization processing; extracting key features from the pre-processed data, wherein the key features include time domain and frequency domain features of physiological parameters, pattern features of behavior activities, periodic features of environmental data, and structured features of medical records; aligning the extracted key features under a unified timestamp and mapping them to a pre-set entity space; based on the aligned feature data, using entity linking and relationship extraction techniques, constructing a multi-modal health knowledge graph with the user as the core node, wherein the node types include user nodes, physiological indicator nodes, symptom nodes, behavior nodes, environment nodes, and disease nodes; the relationship types of the edges include user-has-indicator, indicator-may-trigger-symptom, behavior-affect-indicator, environment-exacerbate-risk, and disease-correlate-symptom; storing and indexing the multi-modal health knowledge graph using a graph database, and dynamically updating the node attributes, edge relationship strengths, and newly added entity relationships in the multi-modal health knowledge graph using an incremental update algorithm.

[0006] Further, the health state assessment and risk prediction module learns embedding representations of nodes and relationships based on the multi-modal health knowledge graph using a graph neural network model, and generates real-time health state assessment results and future short-term health risk prediction results of an individual user in combination with a time series analysis model, and the process includes: based on the multi-modal health knowledge graph, using a graph embedding algorithm to learn low-dimensional vector representations of nodes and relationships in the graph to obtain graph embedding vectors; Based on the time sequence formed by the user historical data, the key features extracted from the preprocessed data are subjected to time sequence analysis, and the time sequence prediction model is used to analyze the change trend of each key feature, so as to obtain the time sequence trend feature of the key feature in the future preset time period; The graph embedding vector and the time sequence trend feature are fused and input into a preset health score prediction model to calculate the current health status score of the user; Based on the extrapolation results of the associated path in the multi-modal health knowledge graph and the time sequence trend feature, the health risk prediction probability of the target health risk event occurring in the future preset time period of the user is predicted, and the key risk influencing factors are identified.

[0007] Further, the process of the target user screening module for determining the target user from all users includes: Based on the health status score and the health risk prediction probability, a comprehensive health risk value is calculated for each user; the comprehensive health risk value is compared with a preset comprehensive health risk threshold; and the user whose comprehensive health risk value exceeds the comprehensive health risk threshold is determined as the target user who needs to be intervened preferentially.

[0008] Further, the process of the individualized intervention strategy generation module for reasoning and generating an individualized intervention sequence for the target user, including a series of ordered intervention actions, according to the health status assessment result and the health risk prediction result, through a dynamic strategy optimization algorithm based on reinforcement learning, includes: The health status assessment result and the health risk prediction result of the user are fused to construct the state in which the reinforcement learning agent is located; A variety of intervention means and their combinations that can be executed are defined as the action space of the agent; A reward function is constructed, which takes the health improvement expectation as positive reward and the risk occurrence probability and user discomfort as negative reward, to guide the learning process of the agent; In the simulation environment composed of the multi-modal health knowledge graph and the user historical data, the agent is trained to learn a strategy that maps the optimal action sequence, i.e. the individualized intervention sequence, according to the current state.

[0009] Further, the process of the intervention execution and feedback adaptation module for converting the individualized intervention sequence into execution instructions matching the smart device interface protocol, driving the connected smart device to execute the intervention, and monitoring the response data of the target user to the intervention in real time, and dynamically adjusting the relationship strength between the intervention strategy and the multi-modal health knowledge graph, includes: The physiological parameters and behavioral activity change data of the user after executing the intervention are collected as response data; The deviation between the response data and the expected improvement target is calculated to evaluate the effectiveness of the intervention strategy; Based on the evaluation results, the parameters of the decision algorithm in the personalized intervention strategy generation module are adjusted using the Bayesian optimization method; According to the intervention effect and the user response, the relationship strength between related entities in the multi-modal health knowledge graph is updated.

[0010] Further, the process of extracting key features from the preprocessed data includes: For physiological parameter data, its time domain features including mean, variance, maximum, and minimum are extracted, and its frequency domain features are extracted through Fourier transform; For behavior activity data, its behavior patterns are identified through clustering algorithm, and the occurrence frequency and duration of different patterns are extracted as pattern features; For environmental data, its daily cycle and weekly cycle features are extracted as periodic features through periodic spectrum analysis; For medical record data, its structured entities and relationships are extracted as structured features through natural language processing technology.

[0011] Further, based on the extrapolation results of the association path and the time series trend features in the multi-modal health knowledge graph, the probability of the occurrence of a target health risk event in a future preset time period is predicted, and the key risk influencing factors are identified. The process includes: In the multi-modal health knowledge graph, the user node corresponding to the target user is located, and is associated to one or more physiological indicator nodes of the current anomaly by the user-has-indicator relationship, as the path starting set; At the same time, the disease node corresponding to the predicted health risk event is located as the path target node; Extract all associated paths from any physiological indicator node in the path starting set to the path target disease node; Calculate the weight of each associated path, where the path weight is determined by the product of the relationship strength of all edges in the path and the risk coefficient of the nodes passed by the path; For each associated path, extract one or more physiological indicator nodes on the path; Directly obtain the time series trend features generated by the time series prediction model for the key features corresponding to these physiological indicator nodes; Fuse the weight of the associated path and the future change value in the time series trend feature, and calculate the final health risk prediction probability through the probability prediction model; Among all the associated paths, the Top-N key paths that contribute most to the final health risk prediction probability are screened out; Extract the nodes connected by the edges with the highest relationship strength on these key paths, where the entities corresponding to the nodes are the key risk influencing factors, which are used to be preferentially targeted in the generated personalized intervention sequence.

[0012] Further, the process of training an agent to learn a strategy includes: The agent selects an action based on its policy network according to the current state; The simulation environment is updated to a new state according to the selected action, and an immediate reward is calculated according to the reward function; The agent updates its policy network parameters according to the state, action, and reward sequence using the policy gradient method; After multiple rounds of iterative training, a stable and optimized policy is finally obtained, which is applied to the user's current health status, and the corresponding personalized intervention sequence is output.

[0013] A data processing method of a smart health system, comprising: S1, collecting physiological parameter data, behavior activity data, environment perception data and medical record data of a user in real time or quasi-real time from multiple heterogeneous data sources; S2, preprocessing, feature extraction and spatio-temporal alignment of the collected multi-source data, and constructing and dynamically updating a multi-modal health knowledge graph based on the processed data; S3, based on the multi-modal health knowledge graph, learning the embedding representation of nodes and relationships using a graph neural network model, and combining a time series analysis model to generate real-time health status evaluation results and future short-term health risk prediction results of the user individual; S4, according to the health status evaluation results and the health risk prediction results, reasoning and generating a personalized intervention sequence for the target user through a dynamic strategy optimization algorithm based on reinforcement learning; S5, converting the personalized intervention sequence into an execution instruction matching the interface protocol of the smart device, driving the connected smart device to execute the intervention, and monitoring the response data of the target user to the intervention in real time; S6, dynamically adjusting the intervention strategy and the relationship strength in the multi-modal health knowledge graph based on the user response data.

[0014] Compared with the prior art, the beneficial effects of the present application are: by collecting data from wearable devices, smart mobile terminals, environmental sensors, hospital information systems and other multi-source heterogeneous data sources in real time, the comprehensiveness and real-time nature of health data are ensured; by preprocessing, feature extraction and alignment of the collected multi-source data, a multi-modal health knowledge graph is dynamically constructed and updated, the scattered data resources are effectively integrated, the data utilization efficiency is improved, and data support is provided for accurate health assessment; based on the multi-modal health knowledge graph, a graph neural network and a time series analysis model are used to realize real-time assessment of user health status and accurate prediction of future health risks, so as to discover health problems in advance and take preventive measures; according to the health assessment results, a personalized intervention sequence is generated through a reinforcement learning algorithm, the pertinence and effectiveness of the intervention are improved, and the positive change of user health behavior is promoted; by converting the intervention strategy into an execution instruction, driving the intelligent device to execute, and monitoring the user response in real time, the intervention strategy is dynamically adjusted, the effective implementation and timely adjustment of the intervention measures are ensured, and the overall effect of health management is improved. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 A module diagram of a smart health system is provided for the present application.

[0016] Figure 2 A flowchart of a data processing method of a smart health system is provided for the present application. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0018] Referring to Figure 1 A smart health system, the system comprising a multi-source data acquisition module, a data fusion and knowledge graph construction module, a health status assessment and risk prediction module, a personalized intervention strategy generation module, and an intervention execution and feedback adaptation module. The multi-source data collection module is used for collecting physiological parameter data, behavior activity data, environment perception data and medical record data of a user in real time or quasi-real time from multiple heterogeneous data sources. In the multi-source data collection module, the heterogeneous data sources include wearable devices, smart mobile terminals, environment sensors and hospital information systems, the physiological parameter data is collected by the wearable devices and includes at least three of heart rate, blood pressure, blood sugar, blood oxygen saturation, body temperature and electrocardiogram waveform; the behavior activity data is collected by the smart mobile terminals and includes sleep duration and quality, step count, exercise type and intensity, sitting time and calorie consumption; the environment perception data is collected by the environment sensors and includes environmental temperature, humidity, air quality index, noise level and light intensity; and the medical record data is obtained through a hospital information system interface and includes diagnosis results, medication records, allergy history and family medical history in an electronic medical record. The data fusion and knowledge graph construction module is used for preprocessing, feature extraction and alignment of the collected multi-source data, and constructing and dynamically updating a multi-modal health knowledge graph based thereon. The nodes represent entities including users, physiological indicators, symptoms, behaviors, environmental factors and diseases, and the edges represent semantic relationships and statistical relationship strengths between entities. The health status assessment and risk prediction module is used for learning the embedding representation of nodes and relationships by using a graph neural network model based on the multi-modal health knowledge graph, and generating real-time health status assessment results and future short-term health risk prediction results of an individual user by combining a time series analysis model. The personalized intervention strategy generation module is used for reasoning and generating a personalized intervention sequence for a target user including a series of ordered intervention actions by using a dynamic strategy optimization algorithm based on reinforcement learning according to the health status assessment results and the health risk prediction results. The intervention execution and feedback adaptation module is used for converting the personalized intervention sequence into execution instructions matching the interface protocol of intelligent devices, driving the connected intelligent devices to execute the intervention, and monitoring the response data of the target user to the intervention in real time, and dynamically adjusting the intervention strategy and the relationship strength of the multi-modal health knowledge graph according to the response data.

[0019] It should be further explained that, in the specific implementation process, the data fusion and knowledge graph construction module preprocesses, extracts features and aligns the collected multi-source data, and constructs and dynamically updates the multi-modal health knowledge graph, including: Preprocessing the multi-source data, including data cleaning, denoising, outlier detection and filling, and format standardization processing; Key features are extracted from the pre-processed data, including time-domain and frequency-domain features of physiological parameters, pattern features of behavioral activities, periodic features of environmental data, and structured features of medical records. Specifically, for physiological parameter data, time-domain features such as mean, variance, maximum, and minimum are extracted, and frequency-domain features are extracted through Fourier transform. For behavioral activity data, clustering algorithms are used to identify behavior patterns, and the frequency and duration of different patterns are extracted as pattern features. For environmental data, daily and weekly periodic features are extracted through periodic pattern analysis. For medical record data, structured entities and relationships are extracted through natural language processing techniques. The extracted key features are aligned under a unified timestamp and mapped to a pre-defined entity space. Based on the aligned feature data, entity linking and relationship extraction techniques are used to construct a multi-modal health knowledge graph with the user as the core node. The node types include user nodes, physiological indicator nodes, symptom nodes, behavior nodes, environment nodes, and disease nodes. The relationship types of edges include user-has-indicator, indicator-may-trigger-symptom, behavior-affects-indicator, environment-aggravates-risk, and disease-associated-symptom. The multi-modal health knowledge graph is stored and indexed using a graph database. As new data continues to flow in, an incremental update algorithm is used to dynamically update the node attributes, edge relationship strength, and new entity relationships in the multi-modal health knowledge graph. Specifically, the incremental update algorithm is a high-efficiency algorithm for processing streaming data. Its core is to update only the affected part of the multi-modal health knowledge graph as new data continues to flow in, thereby reducing computational overhead and improving real-time performance. The specific process includes: A1. When new data flows in through the multi-source data collection module, the system compares the differences between the new data and the existing data in the multi-modal health knowledge graph, identifies new entities, updates node attributes, or changes relationships, such as new heart rate data that may have a deviation from existing data, triggering node attribute updates. A2. For detected changes, related nodes and edges are updated: for node attributes, use sliding window averaging or exponential weighted moving average methods to dynamically adjust attribute values; for edge relationship strength, based on statistical correlation (such as Pearson correlation coefficient) or semantic similarity in new data, recalculate the strength value; for new entity relationships, embed new entities into the graph through entity linking technology and establish edge connections. A3, after updating, re-optimizing the graph structure using the indexing mechanism of the graph database (such as Neo4j) to ensure query efficiency; at the same time, regular consistency checks are performed to avoid loops or contradictions in the graph caused by incremental updates; the incremental update algorithm ensures that the multi-modal health knowledge graph always reflects the latest health status, providing an accurate data foundation for downstream modules.

[0020] It needs to be further explained that, in the specific implementation process, the health status assessment and risk prediction module based on the multi-modal health knowledge graph uses a graph neural network model to learn the embedding representation of nodes and relationships, and combines a time series analysis model to generate real-time health status assessment results and future short-term health risk prediction results of individual users. The process includes: Based on the multi-modal health knowledge graph, a graph embedding algorithm is used to learn the low-dimensional vector representation of each node and relationship in the graph to obtain graph embedding vectors. Specifically, B1, the graph structure based on the multi-modal health knowledge graph, where the node types include user nodes, physiological indicator nodes, symptom nodes, behavior nodes, environment nodes, and disease nodes, and the edge relationship types include user-has-indicator, indicator-may-trigger-symptom, behavior-affect-indicator, environment-aggravate-risk, and disease-correlate-symptom, etc. Each node is associated with key features extracted from preprocessed data, such as: physiological indicator nodes associated with time domain features (such as mean, variance) and frequency domain features (such as Fourier transform coefficients), behavior nodes associated with pattern features (such as behavior pattern frequency generated by clustering), environment nodes associated with periodic features (such as daily and weekly cycles), and medical record nodes associated with structured features (such as entities and relationships). These features serve as the initial feature vectors of the nodes for subsequent embedding learning. B2, a graph neural network model (such as GraphSAGE or Graph Attention Network GAT) is used as the core of the graph embedding algorithm. This model contains a multi-layer message passing mechanism. In each layer, each node updates its representation by aggregating the information of its neighbor nodes. For relationship embedding, the edge relationship type is used as an important parameter in the message passing process, and a relationship-specific weight matrix (for example, in R-GCN) is used to distinguish the influence of different relationship types, thereby simultaneously learning the low-dimensional vector representation of nodes and relationships. B3, the graph neural network model is trained by defining a graph embedding training objective function, which includes node classification loss and link prediction loss. Node classification loss ensures that similar nodes (such as nodes with the same disease type) are close in the embedding space, and link prediction loss ensures that node pairs with strong relationships (such as user-has-indicator) have high inner product scores. During training, the gradient descent algorithm (such as the Adam optimizer) is used to minimize the loss function, and an incremental update mechanism is used to dynamically adjust the embedding representation as new data flows in to reflect the real-time changes in the knowledge graph. B4, output all nodes and relationship low-dimensional vector representation (i.e. graph embedding vector) after training, these vectors capture the semantic relationship between entities in the graph and the statistical correlation strength; the graph embedding vector is then used for health status evaluation module, combined with time series trend characteristics, input into the health score prediction model to generate user health status score and health risk prediction results; In combination with the time series sequence constituted by the user historical data, the key features extracted from the preprocessed data are subjected to time series analysis, and the time series prediction model is used to analyze the change trend of each key feature to obtain the time series trend characteristics of the key features in the future preset time period; The graph embedding vector is combined with the time series trend characteristics and input into the preset health score prediction model to calculate the current health status score of the user; wherein the health score prediction model is one of a linear regression model, a gradient boosting decision tree model or a fully connected neural network model; Based on the extrapolation results of the associated path in the multi-modal health knowledge graph and the time series trend characteristics, the health risk prediction probability of the target health risk event occurring in the future preset time period of the user is predicted, and the key risk influencing factors are identified; Specifically, in the multi-modal health knowledge graph, the user node corresponding to the target user is located, and is associated to one or more physiological indicator nodes of the current abnormality by the user-having-index relationship, as the path starting set; At the same time, the disease node corresponding to the predicted health risk event is located as the path target node; all associated paths from any physiological indicator node in the path starting set to the path target disease node are extracted; The weight of each associated path is calculated, wherein the path weight is determined by the product of the relationship strength of all edges on the path and the risk coefficient of the nodes passed by the path; For each associated path, one or more physiological indicator nodes on the path are extracted; the time series trend characteristics generated by the time series prediction model for the key features corresponding to these physiological indicator nodes are directly obtained; the weight of the associated path and the corresponding future change value in the time series trend characteristics are combined, and the final health risk prediction probability is calculated by a probability prediction model (such as a logistic regression model); wherein the future change value is the specific numerical value of the key feature at a specific time point in the future predicted by the time series trend characteristics; In all the association paths, the top-N key paths that contribute most to the final health risk prediction probability are screened out; the nodes connected by the edges with the highest relationship strength on these key paths are extracted, where the entities corresponding to the nodes are the key risk influencing factors, which are used to be prioritized in the generated personalized intervention sequence; wherein, the extrapolation result refers to using a time series analysis model to perform trend prediction on historical key feature data, thereby inferring the change of feature values in the future time period; specifically, the time series trend feature is analyzed by a time series prediction model (such as ARIMA, LSTM or Prophet model) on the preprocessed key features (such as heart rate, step count) to generate a predicted value sequence for a future preset time period (for example, 24 hours in the future); the extrapolation result includes point prediction (specific numerical value) and interval prediction (confidence interval) for evaluating the health risk change; in risk prediction, the extrapolation result is combined with the association paths of the multi-modal health knowledge graph to quantify the risk probability by calculating the path weight and trend change.

[0021] It needs to be further explained that, in the specific implementation process, the process of the target user screening module for determining the target user from all users includes: Based on the health status score and the health risk prediction probability, a comprehensive health risk value is calculated for each user; wherein, the calculation of the comprehensive health risk value integrates the health status score and the health risk prediction probability into a single index, which is convenient for user screening and priority sorting, and the calculation process is as follows: define the health status score HS as a value of 0-100, which is output based on the health score prediction model, and the lower the value, the worse the health status; define the health risk prediction probability RP as a value of 0-1, which represents the probability of occurrence of a target health risk event (such as a heart attack) in the future; normalize the health status score HS and the health risk prediction probability RP, for example, convert the health status score HS into a relative score (HS'=1-HS / 100), so that the higher the normalized HS' value, the worse the state; calculate the comprehensive health risk value HR using a linear weighting formula: HR=α1×HS'+α2×RP, where α1 and α2 are weight coefficients; compare the comprehensive health risk value with a preset comprehensive health risk threshold; determine the user whose comprehensive health risk value exceeds the comprehensive health risk threshold as a target user who needs to be intervened preferentially; it can be understood that the health status assessment and risk prediction module performs health status assessment and health risk prediction on all users to generate real-time health status assessment results and future short-term health risk prediction results for individual users, where the individual user refers to any evaluated individual in the system, which is all users before screening.

[0022] It needs to be further explained that in the specific implementation process, the process of the personalized intervention strategy generation module reasoning and generating a personalized intervention sequence for the target user, including a series of ordered intervention actions, according to the health state assessment result and the health risk prediction result, through a dynamic strategy optimization algorithm based on reinforcement learning, includes: Fusing the health state assessment result and the health risk prediction result of the user to construct the state in which the reinforcement learning agent is located; Defining a variety of executable intervention means and their combinations as the action space of the agent; Constructing a reward function, which takes the health improvement expectation as positive reward and the risk occurrence probability and user discomfort as negative reward, to guide the learning process of the agent; Training the agent to learn a strategy in a simulated environment composed of a multi-modal health knowledge graph and user historical data, which maps the optimal action sequence, i.e., the personalized intervention sequence, according to the current state; wherein the process of training the agent to learn a strategy includes: The agent selects an action based on its policy network according to the current state; The simulated environment updates to a new state according to the selected action and calculates the immediate reward according to the reward function; in the reinforcement learning training, the specific process of environment update and reward calculation includes: the current state Including the user's health state score , health risk prediction probability and graph embedding vector , i.e. ( represents the current time step, represents the health state score calculated at time step , represents the health risk prediction probability at time step , represents the graph embedding vector at time step ) ; the agent selects an action (such as sending a health reminder or adjusting the drug dosage), the simulated environment updates the state to based on the action and the built-in user model, the user model is constructed using historical data to simulate the real response of the user to the intervention, for example, after performing the action of increasing exercise, the heart rate may decrease and the health score improves; the reward function is: In the formula, represents the change amount of the health state score from the current state to the next state, which provides positive reward when the value is positive (health improvement); a reduction in the probability of health risk, the value is positive (risk reduction) to provide a positive reward; a reduction in the probability of health risk user discomfort or execution cost caused by the execution of the action, quantified by user feedback questionnaire or preset cost table (for example, the cost of high-intensity intervention is higher than that of reminder intervention), which is a negative reward; a predefined weight coefficient for balancing the relative importance of health improvement, risk reduction and user comfort; the process is repeated in the simulation environment, and the agent learns to maximize the cumulative reward through the policy gradient method, and finally outputs the optimized strategy; The agent uses the policy gradient method to update its policy network parameters according to the state, action and reward sequence; After multiple rounds of iterative training, a stable and optimized strategy is finally obtained, which is applied to the current health status of the user, and the corresponding personalized intervention sequence is output.

[0023] It should be further explained that in the specific implementation process, the intervention execution and feedback adaptation module converts the personalized intervention sequence into execution instructions matching the intelligent device interface protocol, drives the connected intelligent device to execute the intervention, and real-time monitors the response data of the target user to the intervention, and the process of dynamically adjusting the relationship strength between the intervention strategy and the multi-modal health knowledge graph based on the response data includes: Collect the physiological parameters and behavior activity change data of the user after executing the intervention as response data; Based on the expected health status improvement direction of the personalized intervention sequence, set specific and quantifiable expected improvement targets; calculate the deviation between the response data and the expected improvement targets to evaluate the effectiveness of the intervention strategy; Based on the evaluation results, the Bayesian optimization method is used to adjust the parameters of the decision algorithm in the personalized intervention strategy generation module; wherein the Bayesian optimization method is used to adjust the parameters (such as learning rate, discount factor) of the reinforcement learning algorithm in the personalized intervention strategy generation module to improve the intervention effect, and the specific process includes: F1, define the parameter set to be optimized, mainly including the weight coefficients of each item in the reinforcement learning agent reward function, such as the weight of health improvement expectation, risk occurrence probability and user discomfort; construct a Bayesian optimization objective function with intervention effect as the measurement standard, wherein the function is calculated based on the deviation between the user response data (such as physiological parameters, actual changes in behavior activities) and the preset quantifiable expected improvement target; F2, Bayesian optimization iteration is performed: in the initialization stage, a set of initial parameter values is randomly selected from the parameter space; a Gaussian process is used as a surrogate model to probabilistically model the objective function; an expected improvement function is used to balance exploration and exploitation and select the next set of parameters to be evaluated; in each iteration, the current parameters are used to run the decision algorithm to generate and execute the intervention, collect user response data to calculate the objective function value, and update the surrogate model accordingly; when the objective function value converges or the maximum number of iterations is reached, the optimization is terminated and the optimal parameter combination found is output; finally, the optimal parameters obtained by optimization are applied to the decision algorithm (i.e., the reinforcement learning agent) in the personalized intervention strategy generation module to generate more effective subsequent intervention strategies; According to the intervention effect and the user response, the relationship strength between related entities in the multi-modal health knowledge graph is updated, the association between effective intervention and positive results is enhanced, and the association of ineffective intervention is weakened.

[0024] Reference Figure 2 A data processing method of a smart health system, comprising: S1, real-time or quasi-real-time acquisition of physiological parameter data, behavior activity data, environmental perception data and medical record data of users from multiple heterogeneous data sources; S2, pre-processing, feature extraction and spatio-temporal alignment of the collected multi-source data, and construction and dynamic updating of a multi-modal health knowledge graph based on the processed data; S3, based on the multi-modal health knowledge graph, learning the embedding representation of nodes and relationships using a graph neural network model, and combining a time series analysis model to generate real-time health status assessment results and future short-term health risk prediction results of individual users; S4, according to the health status assessment results and the health risk prediction results, a dynamic strategy optimization algorithm based on reinforcement learning is used to infer and generate a personalized intervention sequence for the target user; S5, converting the personalized intervention sequence into execution instructions compatible with the interface protocol of the intelligent device to drive the connected intelligent device to execute the intervention, and monitoring the response data of the target user to the intervention in real time; S6, based on the user response data, dynamically adjusting the intervention strategy and the relationship strength in the multi-modal health knowledge graph.

[0025] In addition, the formula involved in the above is calculated by removing the dimension and taking the numerical value, is obtained by collecting a large amount of data to obtain a formula closest to the real situation, and the weight coefficient in the formula and the preset threshold in the analysis process are set by the person skilled in the art according to the actual situation or obtained by a large amount of data simulation; The size of the weight coefficient is a specific value obtained by quantifying each parameter, which is convenient for subsequent comparison. The size of the weight coefficient depends on the number of sample data and the corresponding processing coefficient preliminarily set by the person skilled in the art for each group of sample data; As long as it does not affect the proportional relationship between the parameters and the quantized values.

[0026] Each embodiment in the specification is described in a progressive manner, and the same and similar parts between each embodiment can be referred to each other. Each embodiment mainly describes the difference from other embodiments. In particular, for the device embodiment, since it basically relies on the method embodiment, the description is relatively simple, and the related parts can be referred to the part of the method embodiment.

[0027] In order to describe conveniently, the above device is described by dividing into various units according to functions. Of course, the functions of each unit can be implemented in the same or multiple software and / or hardware in the implementation of the present application.

[0028] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt the form of a computer program product implemented on one or more computer usable storage media (including but not limited to disk memory, CD-ROM, optical memory, etc.) containing computer usable program code.

[0029] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 One or more flows and / or blocks Figure 1 An apparatus that performs the functions specified in one or more flows and / or blocks.

[0030] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 The functions specified in the flow or flows and / or blocks Figure 1 The functions specified in the flow or flows and / or blocks

[0031] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions that are executed on the computer or other programmable apparatus provide steps for implementing the functions specified in the flow Figure 1 The functions specified in the flow or flows and / or blocks Figure 1 The functions specified in the flow or flows and / or blocks

[0032] Secondly, the drawings of the embodiments disclosed in the present application only involve the structures related to the embodiments disclosed in the present application, and other structures can refer to the general design, and the same embodiments and different embodiments of the present application can be combined with each other under the condition of no conflict; Finally, the above description is only the preferred specific implementation of the present application, but the protection scope of the present application is not limited to this, any person skilled in the art can make equivalent replacement or change according to the technical scheme and the inventive concept of the present application within the technical range disclosed in the present application, which should be covered in the protection scope of the present application.

Claims

1. A smart health system, characterized in that, The system comprises a multi-source data acquisition module, a data fusion and knowledge graph construction module, a health status evaluation and risk prediction module, a personalized intervention strategy generation module, and an intervention execution and feedback adaptation module. The multi-source data acquisition module is configured to acquire physiological parameter data, behavior activity data, environmental perception data, and medical record data of a user in real time or quasi-real time from multiple heterogeneous data sources. The data fusion and knowledge graph construction module is configured to preprocess, extract, and align the acquired multi-source data, and construct and dynamically update a multi-modal health knowledge graph based on the preprocessed, extracted, and aligned data. The health status evaluation and risk prediction module is configured to learn the embedding representation of nodes and relationships by using a graph neural network model based on the multi-modal health knowledge graph, and generate real-time health status evaluation results and future short-term health risk prediction results of the user by combining a time series analysis model. The personalized intervention strategy generation module is configured to infer and generate a personalized intervention sequence for the target user by using a dynamic strategy optimization algorithm based on reinforcement learning according to the health status evaluation results and the health risk prediction results. The intervention execution and feedback adaptation module is configured to convert the personalized intervention sequence into execution instructions matching the interface protocol of the intelligent device, drive the connected intelligent device to execute the intervention, and dynamically adjust the intervention strategy and the relationship strength of the multi-modal health knowledge graph according to the response data of the target user to the intervention. 2.The smart health system of claim 1, wherein, The data fusion and knowledge graph construction module preprocesses, extracts, and aligns the acquired multi-source data, and constructs and dynamically updates the multi-modal health knowledge graph. The preprocessing of the multi-source data includes data cleaning, denoising, outlier detection and filling, and format standardization processing. The key features extracted from the preprocessed data include time domain and frequency domain features of physiological parameters, pattern features of behavior activities, periodic features of environmental data, and structured features of medical records. The extracted key features are aligned under a unified timestamp and mapped to a preset entity space. Based on the aligned feature data, entity linking and relationship extraction techniques are used to construct a multi-modal health knowledge graph with the user as the core node, wherein the node types include user nodes, physiological indicator nodes, symptom nodes, behavior nodes, environment nodes, and disease nodes; and the relationship types of edges include user-has-indicator, indicator-may-trigger-symptom, behavior-affect-indicator, environment-aggravate-risk, and disease-correlate-symptom. The multi-modal health knowledge graph is stored and indexed by using a graph database, and the node attributes, relationship strengths of edges, and newly added entity relationships in the multi-modal health knowledge graph are dynamically updated by using an incremental update algorithm. 3.The smart health system of claim 1, wherein, The health status evaluation and risk prediction module learns embedding representations of nodes and relationships based on the multi-modal health knowledge graph using a graph neural network model, and generates real-time health status evaluation results and future short-term health risk prediction results of the individual user by combining a time series analysis model. The process includes: Based on the multi-modal health knowledge graph, the graph embedding algorithm is used to learn the low-dimensional vector representation of each node and relationship in the graph to obtain the graph embedding vector. The time series sequence composed of user historical data is used to perform time series analysis on the key features extracted from the preprocessed data, and a time series prediction model is used to analyze the trend of each key feature to obtain the time series trend feature of the key feature in the future preset time period. The graph embedding vector and the time series trend feature are fused and input into the preset health score prediction model to calculate the current health status score of the user. Based on the correlation path in the multi-modal health knowledge graph and the extrapolation result of the time series trend feature, the health risk prediction probability of the target health risk event occurring in the future preset time period is predicted, and the key risk influencing factors are identified. 4.The smart health system of claim 3, wherein, The system also includes a target user screening module for determining the target user from all users. The process includes: Based on the health status score and the health risk prediction probability, a comprehensive health risk value is calculated for each user. The comprehensive health risk value is compared with the preset comprehensive health risk threshold. The user whose comprehensive health risk value exceeds the comprehensive health risk threshold is determined as the target user who needs to be intervened preferentially. 5.The smart health system of claim 1 or 4, wherein, The personalized intervention strategy generation module reasons and generates a personalized intervention sequence for the target user, including a series of ordered intervention actions, based on the health status evaluation result and the health risk prediction result, through a dynamic strategy optimization algorithm based on reinforcement learning. The process includes: The health status evaluation result and the health risk prediction result of the user are fused to construct the state in which the reinforcement learning agent is located. A variety of executable intervention methods and their combinations are defined as the action space of the agent. A reward function is constructed, which takes the health improvement expectation as positive reward and the risk occurrence probability and user discomfort as negative reward, to guide the learning process of the agent. In the simulated environment composed of the multi-modal health knowledge graph and the user historical data, the agent is trained to learn a strategy that maps the current state to the optimal action sequence, i.e. the personalized intervention sequence. 6.The smart health system of claim 1, wherein, The intervention execution and feedback adaptation module converts the personalized intervention sequence into execution instructions compatible with the smart device interface protocol, drives the connected smart device to execute the intervention, and monitors the response data of the target user to the intervention in real time. The process includes: The physiological parameters and behavioral activity change data of the user after the intervention are collected as response data. The deviation between the response data and the expected improvement target is calculated to evaluate the effectiveness of the intervention strategy. Based on the evaluation result, the Bayesian optimization method is used to adjust the parameters of the decision algorithm in the personalized intervention strategy generation module. According to the intervention effect and the user response, the relationship strength between related entities in the multi-modal health knowledge graph is updated. 7.The smart health system of claim 2, wherein, The process of extracting key features from pre-processed data includes: For physiological parameter data, extracting its time domain features including mean, variance, maximum, minimum, and extracting its frequency domain features through Fourier transform; For behavior activity data, identifying its behavior patterns through clustering algorithm, and extracting the frequency and duration of different patterns as pattern features; For environmental data, extracting its daily cycle, weekly cycle features as cycle features through periodic pattern analysis; For medical record data, extracting its structured entities and relationships as structured features through natural language processing technology. 8.The smart health system of claim 3, wherein, Based on the extrapolation results of the association path and the time series trend features in the multi-modal health knowledge graph, the process of predicting the probability of the occurrence of a target health risk event in a future preset time period for a user and identifying the key risk influencing factors includes: In the multi-modal health knowledge graph, locate the user node corresponding to the target user, and associate it to one or more physiological indicator nodes with current abnormalities mapped by key features through user-has-indicator relationship, as the path starting set; At the same time, locate the disease node corresponding to the predicted health risk event as the path target node; Extract all associated paths from any physiological indicator node in the path starting set to the path target disease node; Calculate the weight of each associated path, where the path weight is determined by the product of the relationship strength of all edges in the path and the risk coefficient of the nodes passed by the path; For each associated path, extract one or more physiological indicator nodes on the path; directly obtain the time series trend features generated by the time series prediction model for the key features corresponding to these physiological indicator nodes; fuse the weight of the associated path and the future change value in the time series trend feature, and calculate the final health risk prediction probability through the probability prediction model; In all associated paths, filter out the Top-N key paths that contribute most to the final health risk prediction probability; Extract the nodes connected by the edges with the highest relationship strength on these key paths, where the entity corresponding to the node is the key risk influencing factor, which is used to be prioritized in the generated personalized intervention sequence. 9.The smart health system of claim 5, wherein, The process of training an agent to learn a strategy includes: The agent selects an action based on its policy network according to the current state; The simulation environment updates to a new state according to the selected action, and calculates the immediate reward according to the reward function; The agent updates its policy network parameters according to the state, action, and reward sequence using the policy gradient method; After multiple rounds of iterative training, a stable and optimized strategy is finally obtained, which is applied to the user's current health status, and the corresponding personalized intervention sequence is output.

10. A data processing method of a smart health system, characterized in that, Applied to a smart health system as claimed in any one of claims 1-9, the method comprises: S1, real-time or quasi-real-time acquisition of physiological parameter data, behavior activity data, environmental perception data and medical record data of a user from multiple heterogeneous data sources; S2, pre-processing, feature extraction and spatio-temporal alignment of the collected multi-source data, and constructing and dynamically updating a multi-modal health knowledge graph based on the processed data; S3, based on the multi-modal health knowledge graph, the graph neural network model is used to learn the embedding representation of nodes and relationships, and combined with the time series analysis model, the real-time health status evaluation result and the future short-term health risk prediction result of the individual user are generated; S4, according to the health status evaluation result and the health risk prediction result, through the dynamic strategy optimization algorithm based on reinforcement learning, the individual intervention sequence for the target user is reasoned and generated; S5, the individual intervention sequence is converted into an execution instruction matched with the intelligent device interface protocol, and the connected intelligent device is driven to execute the intervention, and the response data of the target user to the intervention is monitored in real time; S6, based on the user response data, the intervention strategy and the relationship strength in the multi-modal health knowledge graph are dynamically adjusted.

Images