Household intelligent health management system and method thereof

Abstract

The invention relates to the technical field of intelligent medical treatment, in particular to a household intelligent health management system and a method thereof.The system collects household environment parameters and user physiological indexes through a data collection subsystem, constructs a medical knowledge graph of a three-layer structure through a knowledge graph driving subsystem, and dynamically updates the medical knowledge graph; the multi-target health intervention subsystem generates a candidate intervention scheme set based on a knowledge graph, the digital twinborn verification subsystem predicts and selects a scheme with a health improvement index higher than a target by establishing a user, environment and interactive twinborn model, and the adaptive interaction subsystem pushes the verified scheme to a user in a personalized manner and collects feedback information. Meanwhile, the digital twinborn verification subsystem transmits the feedback information to the knowledge graph driving subsystem to trigger dynamic updating of the knowledge graph, the system comprehensively collects health influence factors, comprehensive data support is provided for health risk assessment, and personalized health management is achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent medical technology, and in particular to a home intelligent health management system and method thereof. Background Art

[0002] With the increasing incidence of chronic diseases and the aging population, the demand for family health management is becoming increasingly urgent. Currently, there are many health monitoring and management systems on the market, but these systems generally have the following shortcomings: First, existing systems are mostly based on single data collection, such as monitoring only physiological indicators or environmental parameters. They lack comprehensive analysis of multi-source heterogeneous data, making it difficult to comprehensively assess health risks. Secondly, traditional health intervention plans are often generated based on simple empirical rules, with limited personalization and unable to meet the specific needs of different users. Furthermore, existing systems lack predictive verification mechanisms for intervention effects, making it difficult for users to predict the actual effects of intervention plans, thereby affecting intervention compliance. In addition, most systems use a fixed knowledge base and are unable to dynamically update medical knowledge based on user feedback, making it difficult to adapt to the rapid development of medical knowledge. Finally, the existing system's single interaction method fails to fully consider the language and cognitive differences of different user groups, resulting in a poor user experience.

[0003] Therefore, there is an urgent need for an interactive and friendly home smart health management system that can integrate multi-source data, provide highly personalized intervention plans, have predictive verification capabilities, support dynamic knowledge updates, and be interactive. Summary of the Invention

[0004] The purpose of the present invention is to provide a home intelligent health management system and a personalized intervention method thereof, aiming to solve the problems existing in the prior art, realize the correlation analysis between environmental factors and personal health, personalized intervention based on medical knowledge graph, predictive verification of intervention effects and closed-loop optimization.

[0005] The present invention proposes a home intelligent health management system, comprising: The data acquisition subsystem includes an environmental monitoring module and a physiological monitoring module, which are used to collect home environment parameters and user physiological indicators; A knowledge graph driving subsystem, connected to the data acquisition subsystem, is configured to receive the environmental parameters and the user's physiological indicators, construct a three-layer medical knowledge graph comprising an entity layer, a relationship layer, and a knowledge layer, and dynamically update the medical knowledge graph based on preset update trigger conditions; A multi-objective health intervention subsystem, connected to the knowledge graph driving subsystem, is used to generate a set of candidate intervention plans based on the medical knowledge graph by setting a multi-objective function of nutritional balance goals, energy balance goals, and medication compliance goals; A digital twin verification subsystem, connected to the multi-objective health intervention subsystem, is used to establish a user twin model, an environment twin model, and an interaction twin model, perform predictive verification on the candidate intervention scheme set, and select the scheme with the predicted health improvement index higher than the target threshold as the verified scheme; An adaptive interaction subsystem, connected to the digital twin verification subsystem, is configured to receive the verification plan, push the verification plan to the user through a personalized push strategy based on multimodal recognition of medical terms and user profile analysis, and collect user feedback information; Among them, the digital twin verification subsystem is also connected to the knowledge graph driving subsystem to transmit the user feedback information to the knowledge graph driving subsystem to trigger the dynamic update of the medical knowledge graph.

[0006] Preferably, the knowledge graph driven subsystem includes: Entity recognition module, which converts entity names in medical literature into standard medical identifiers; A three-layer structure building module is used to build a medical knowledge graph consisting of an entity layer, a relationship layer, and a knowledge layer. The entity layer includes drugs, symptoms, diseases, departments, equipment, ingredients, acupoints, and action entities; the relationship layer includes attribute relationships and inter-entity relationships; and the knowledge layer includes diagnostic rules, treatment plans, and health recommendations. Adaptive update module, used to perform entity update, relationship update, attribute update and data update when the preset update trigger conditions are met; The knowledge query module is used to retrieve relevant medical knowledge based on the user's health status and needs, and support the generation of intervention plans.

[0007] Preferably, the environmental monitoring module includes: High-frequency sampling unit, used to collect home environment data every 30 seconds through PM2.5 sensor and temperature sensor; A convolution feature extraction unit, configured to perform convolution and pooling processing on the environmental data to extract local environmental features; Environmental grading assessment unit, used to compare the extracted environmental characteristics with preset health standards and classify the environmental status into three levels: low risk, medium risk and high risk; The early warning trigger unit is used to generate early warning information when the environmental parameters exceed the preset threshold value and incorporate the early warning information into the intervention plan.

[0008] Preferably, the physiological monitoring module includes: Multi-source data acquisition unit, used to collect users' physiological indicators such as heart rate, blood pressure, blood oxygen, body temperature, as well as behavioral data such as sleep, activity, and stress through smart bracelets and smartphones; a time series feature extraction unit, configured to process the physiological indicators and the behavioral data through a time series feature extraction network to capture time dependencies and important features; The health record construction unit is used to construct a health record that records the user's health baseline, fluctuation range and abnormal pattern based on the processed physiological data.

[0009] Preferably, the multi-target health intervention subsystem includes: Objective function setting module, used to set multi-objective functions including nutritional balance target, energy balance target and medication compliance target; Constraint setting module, used to set time constraints, health impact constraints, nutritional balance constraints and energy consumption constraints; An optimization algorithm module, configured to perform non-dominated sorting on candidate solutions based on the multi-objective function and the constraints, and identify a Pareto optimal solution set; The decision matrix module is used to construct the decision matrix of candidate solutions, calculate the comprehensive score through normalization processing and weighted evaluation, and select the solution with the highest score as the final intervention solution.

[0010] Preferably, the digital twin verification subsystem includes: User twin model construction module, used to build a virtual user model based on user attributes such as age, gender, weight, and underlying diseases; The environmental twin model construction module is used to build a virtual environment model based on the layout, temperature, humidity, air quality and other parameters of the home environment; The interactive twin model construction module is used to build an interactive model that describes the user's behavior patterns and health responses under different environmental conditions based on the interaction data between the user and the environment; The time series simulation module is used to simulate a one-week intervention cycle and sample user health indicators at six key time points every day; The health status prediction module is used to predict the changes in the user's health status after the intervention cycle based on the user's initial health status and intervention behavior.

[0011] Preferably, the adaptive interaction subsystem includes: Medical term recognition module, which is used to identify medical terms in user input through medical dictionary matching, context analysis, and model fusion; The dialect adaptation module is used to extract the unique acoustic and semantic features of the dialect and realize the conversion between the dialect and Mandarin; Multi-language support module, used to provide real-time translation function in multiple languages; A personalized push module, which optimizes push timing, adjusts content granularity, and personalizes presentation based on user activity patterns, attention patterns, expertise, and preferences. The interaction frequency adjustment module is used to monitor the user's response pattern to push information and dynamically adjust the push frequency to avoid user fatigue and resistance.

[0012] Preferably, the update triggering conditions include: the nutritional balance of the intervention plan reaches more than 90%, the calorie error of the exercise plan is within the range of plus or minus 5%, and the medication compliance reaches 100%.

[0013] Preferably, the home smart health management system supports three deployment modes: basic version, standard version and advanced version, among which: The basic version implements environmental monitoring, physiological monitoring and basic health advice functions; The standard version adds a knowledge graph driven subsystem and a multi-target health intervention subsystem based on the basic version. The advanced version adds a digital twin verification subsystem and multimodal recognition function of medical terms based on the functions of the standard version.

[0014] Home smart health management methods, including: Collect home environment parameters and user physiological indicators; Based on the environmental parameters and the user's physiological indicators, a three-layer medical knowledge graph is constructed, comprising an entity layer, a relationship layer, and a knowledge layer; Based on the medical knowledge graph, a set of candidate intervention plans is generated by setting a multi-objective function of nutritional balance goals, energy balance goals, and medication compliance goals; Establish user twin models, environment twin models, and interaction twin models, perform predictive verification on the candidate intervention plan set, and select plans with predicted health improvement indicators above the target threshold as verified plans; Based on multimodal recognition of medical terms and user portrait analysis, the verification solution is pushed to the user through a personalized push strategy; Collect user feedback information and trigger the dynamic update of the medical knowledge graph when the preset update trigger conditions are met; Among them, the update trigger conditions include: the nutritional balance of the intervention plan reaches more than 90%, the calorie error of the exercise plan is within the range of plus or minus 5%, and the medication compliance reaches 100%.

[0015] The beneficial effects of the present invention include: 1. By integrating environmental monitoring and physiological monitoring, comprehensive collection of health-influencing factors is achieved, providing comprehensive data support for health risk assessment; 2. The three-layer medical knowledge graph, combined with a dynamic update mechanism, ensures the professionalism and timeliness of health interventions, significantly improving the scientific nature of intervention plans; 3. The intervention plan generation mechanism based on a multi-objective optimization algorithm balances multiple goals such as nutritional balance, energy balance, and medication compliance, achieving highly personalized intervention plans and increasing the implementation rate by 65%; 4. Innovatively introduce digital twin technology to establish a predictive verification mechanism, predicting and adjusting possible adverse effects before intervention is implemented, increasing the accuracy of intervention effect prediction to 78%; 5. Through adaptive interaction and personalized push strategies, the system has improved its inclusiveness and ease of use for different user groups, and user satisfaction is 40% higher than that of traditional systems; 6. A complete closed loop from data collection to intervention feedback has been built, achieving continuous optimization of health interventions and improving health indicators by 35% compared to traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is the overall architecture diagram of the family smart health management system of the present invention; Figure 2 It is a schematic diagram of the structure of the knowledge graph driven subsystem of the present invention; Figure 3 is a flow chart of the multi-target health intervention subsystem of the present invention; Figure 4 This is a working principle diagram of the digital twin verification subsystem of the present invention; Figure 5 It is a functional module diagram of the adaptive interaction subsystem of the present invention; Figure 6 It is a flow chart of the family intelligent health management method of the present invention. DETAILED DESCRIPTION

[0017] Please refer to the attached Figure 1-6 , the specific implementation of the present invention is further described in detail below with reference to the accompanying drawings.

[0018] like Figure 1 As shown, the home intelligent health management system provided by the present invention includes a data acquisition subsystem 1, a knowledge graph driving subsystem 2, a multi-target health intervention subsystem 3, a digital twin verification subsystem 4 and an adaptive interaction subsystem 5.

[0019] The data acquisition subsystem 1 includes an environmental monitoring module 11 and a physiological monitoring module 12, which are used to collect home environmental parameters and user physiological indicators. Preferably, the environmental parameters monitored by the environmental monitoring module 11 mainly include PM2.5 concentration, indoor temperature, humidity, noise, etc.; the user physiological indicators monitored by the physiological monitoring module 12 mainly include heart rate, blood pressure, blood oxygen, body temperature, etc., as well as behavioral data such as sleep, activity, and stress.

[0020] The knowledge graph driver subsystem 2 is connected to the data acquisition subsystem 1 and is used to receive environmental parameters and user physiological indicators, construct a three-layer medical knowledge graph consisting of an entity layer, a relationship layer, and a knowledge layer, and dynamically update the medical knowledge graph based on preset update trigger conditions. In one embodiment of the present invention, the entity layer of the three-layer medical knowledge graph includes eight core entities: drugs, symptoms, diseases, departments, equipment, ingredients, acupoints, and actions; the relationship layer includes attribute relationships (such as the relationship between departments and diseases) and inter-entity relationships (such as the relationship between diseases); and the knowledge layer includes advanced knowledge such as diagnostic rules, treatment plans, and health recommendations.

[0021] The multi-objective health intervention subsystem 3 is connected to the knowledge graph driven subsystem 2 and is used to generate a set of candidate intervention plans based on the medical knowledge graph by setting a multi-objective function of nutritional balance goals, energy balance goals, and medication compliance goals. In an embodiment of the present invention, the nutritional balance goal is set to a nutritional balance of more than 90%, the energy balance goal is set to an error between the actual calorie consumption and the target calorie consumption within ±5%, and the medication compliance goal is set to 100% medication compliance.

[0022] Digital twin verification subsystem 4 is connected to the multi-objective health intervention subsystem 3 to establish user twin models, environment twin models, and interaction twin models. It then performs predictive verification on a set of candidate intervention plans and selects those plans whose predicted health improvement indicators exceed the target threshold as the ones that pass verification. Furthermore, digital twin verification subsystem 4 is connected to the knowledge graph driver subsystem 2 to transmit user feedback to the knowledge graph driver subsystem 2, triggering dynamic updates to the medical knowledge graph.

[0023] The adaptive interaction subsystem 5 is connected to the digital twin verification subsystem 4 and is used to receive verification solutions. Based on multimodal recognition of medical terminology and user profile analysis, it pushes the verification solutions to users through a personalized push strategy and collects user feedback. Preferably, the personalized push strategy includes push timing optimization, content granularity adjustment, and personalized presentation to improve user acceptance and execution rates.

[0024] like Figure 2As shown, in a preferred embodiment of the present invention, the knowledge graph driving subsystem 2 includes an entity recognition module 21, a three-layer structure construction module 22, an adaptive update module 23 and a knowledge query module 24.

[0025] The entity recognition module 21 is used to convert entity names in medical literature into standard medical identifiers. The present invention adopts an improved named entity recognition technology to process the entity recognition task as a sequence labeling problem. Specifically, the module first inputs the medical text X={x1,x2,...,x n}, where x_i represents the i-th word, and then a multi-layer neural network is used to generate the context representation H={h1,h2,...,h n}, and finally predict the label Y={y1,y2,...,y n}, enabling the recognition and classification of medical entities. In a family health management scenario, when a user describes symptoms such as "recent dizziness, headache, and high blood pressure," the system can accurately identify "dizziness" and "headache" as symptom entities, and "high blood pressure" as a hypertension disease entity, and then associate corresponding health recommendations.

[0026] The three-layer structure building module 22 is used to build a medical knowledge graph that includes an entity layer, a relationship layer, and a knowledge layer. In the entity layer, each entity contains a unique identifier, a type, a name, and a set of attributes. For example, the data structure of the entity "hypertension" is: { "id":"E10086", "type":"Disease", "name":"hypertension", "attributes":{ "ICD-10":"I10", "risk_level":3, "chronic":true } } In the relationship layer, each relationship contains the relationship type, source entity, target entity, and relationship attributes. For example, the relationship structure between hypertension and headache is: { "id":"R5042", "type":"has_symptom", "source":"E10086", / / High blood pressure "target":"E20045", / / headache "attributes":{ "confidence":0.85, "frequency":"common", "correlation_type":"indicative" } } The knowledge layer includes high-level knowledge units such as diagnostic rules, treatment plans, and health recommendations, forming a complete medical knowledge system. For example, for patients with hypertension, the knowledge layer might include health recommendations such as "limit sodium intake to no more than 5g / day" and "walk for 30 minutes daily." This knowledge serves as the basis for generating personalized intervention plans.

[0027] The adaptive update module 23 is used to perform entity updates, relationship updates, attribute updates, and data updates when preset update trigger conditions are met. The update trigger conditions of the present invention are: the nutritional balance of the intervention plan reaches above 90%, the calorie error of the exercise plan is within ±5%, and the medication compliance reaches 100%. When these conditions are met, the system will update the entities, relationships, and attributes in the knowledge graph based on user feedback data and the latest medical knowledge to ensure the timeliness and accuracy of medical knowledge. For example, when most hypertensive patients experience significant improvement in blood pressure after implementing the "walk 30 minutes a day" recommendation, the system will strengthen the association strength of this recommendation with hypertension control and increase the weight of this recommendation in future plans.

[0028] The knowledge query module 24 is used to retrieve relevant medical knowledge based on the user's health status and needs, supporting the generation of intervention plans. This module supports multi-dimensional semantic queries based on entities, relationships, and attributes. It can retrieve relevant diagnostic rules, treatment plans, and health recommendations based on the user's health status, providing knowledge support for the generation of intervention plans. In actual application, when the system detects that the user's blood pressure is elevated, the knowledge query module will retrieve all health recommendations related to hypertension and select the most appropriate recommendations based on the user's specific situation (such as age, complications, lifestyle habits, etc.) as the basis for the intervention plan.

[0029] like Figure 1 As shown, in a preferred embodiment of the present invention, the environment monitoring module 11 includes a high-frequency sampling unit 111 , a convolution feature extraction unit 112 , an environment classification assessment unit 113 and an early warning triggering unit 114 .

[0030] The high-frequency sampling unit 111 is used to collect home environment data every 30 seconds through the PM2.5 sensor and temperature sensor. The selection of a 30-second sampling interval is based on research on the rate of change of the home environment, which can ensure real-time performance while avoiding the generation of excessive redundant data. In a home environment, air quality and temperature usually do not change drastically in a short period of time. The 30-second interval is sufficient to capture environmental change trends and can also respond to sudden changes in a timely manner, such as a rapid drop in temperature caused by opening windows for ventilation or an increase in indoor PM2.5 concentration caused by cooking.

[0031] The convolution feature extraction unit 112 is used to perform convolution and pooling processing on the environmental data to extract local environmental features. This unit uses a multi-layer convolutional network structure to process the environmental data matrix. The specific structure includes: input layer (environmental data matrix 96×96×3), convolution layer 1 (32 3×3 convolution kernels, activation function ReLU), pooling layer 1 (2×2 maximum pooling), convolution layer 2 (64 3×3 convolution kernels, activation function ReLU), pooling layer 2 (2×2 maximum pooling), fully connected layer 1 (512 neurons, activation function ReLU), fully connected layer 2 (5 neurons, corresponding to environmental classification). In practical applications, the environmental data matrix contains spatial distribution information, such as environmental parameters in different areas such as the living room, bedroom, and kitchen. Convolutional feature extraction can effectively identify environmental change patterns in specific areas, such as increased local PM2.5 concentrations caused by cooking in the kitchen.

[0032] The environmental classification assessment unit 113 is used to compare the extracted environmental characteristics with the preset health standards and classify the environmental status into three levels: low risk, medium risk and high risk. The specific standards are: PM2.5 concentration <35μg / m 3 Low risk, 35-75 μg / m 3 Intermediate risk, >75 μg / m 3 The indoor temperature is 18-26℃ for low risk, 15-18℃ or 26-30℃ for medium risk, and <15℃ or >30℃ for high risk; the indoor humidity is 40-60% for low risk, 30-40% or 60-70% for medium risk, and <30% or >70% for high risk. These thresholds are based on the World Health Organization (WHO) and national environmental protection standards, combined with human comfort and health impact research. For example, PM2.5 concentrations exceeding 75μg / m 3 It will significantly increase the risk of respiratory diseases, especially for the elderly and children; room temperature below 15℃ or above 30℃ exceeds the human comfort range, which may cause blood vessels to constrict or dilate and affect blood pressure stability.

[0033] The warning trigger unit 114 is used to generate warning information when the environmental parameters exceed the preset threshold and incorporate the warning information into the intervention plan. For example, when the PM2.5 concentration exceeds 75μg / m3 If the indoor temperature drops below 15°C, the system will generate a low-temperature warning and recommend users to adjust the heating or wear warm clothes. These warnings and recommendations are directly related to the user's health status. For example, for asthma patients, the system will generate an air quality warning when the PM2.5 concentration reaches 50μg / m 3 For patients with hypertension, the system will pay more attention to changes in room temperature, because temperature fluctuations may cause blood pressure fluctuations.

[0034] like Figure 1 As shown, in a preferred embodiment of the present invention, the physiological monitoring module 12 includes a multi-source data acquisition unit 121 , a time series feature extraction unit 122 and a health record construction unit 123 .

[0035] The multi-source data acquisition unit 121 is used to collect physiological indicators such as the user's heart rate, blood pressure, blood oxygen, and body temperature, as well as behavioral data such as sleep, activity, and stress through smart bracelets and smartphones. Specifically, the heart rate sampling frequency is 1Hz, the blood pressure sampling frequency is 30 minutes / time, the blood oxygen sampling frequency is 1Hz, and the body temperature sampling frequency is 30 minutes / time. The setting of these sampling frequencies is based on medical research results and can reduce device energy consumption and user burden while ensuring data validity. In the home health management scenario, real-time monitoring of heart rate and blood oxygen (1Hz) is particularly important for patients with cardiovascular disease, as abnormal fluctuations can be detected in a timely manner; blood pressure and body temperature are relatively stable, and a sampling frequency of 30 minutes / time is sufficient to track daily changes. For special user groups, such as heart patients, the system also supports the collection of electrocardiogram data (sampling frequency 250Hz) to more accurately monitor symptoms such as arrhythmia.

[0036] The time series feature extraction unit 122 is used to process physiological indicators and behavioral data using a time series feature extraction network, capturing temporal dependencies and important features. This unit utilizes a time series feature extraction network, which comprises a feature extraction layer, memory units, and an attention mechanism, effectively processing high-dimensional time series physiological signals. When processing heart rate data, the system first segments the raw heart rate data into 5-minute windows. It then calculates features such as the mean (HR_mean), maximum (HR_max), minimum (HR_min), standard deviation (HR_std), and number of peaks (HR_peaks) within each window. The network then extracts the temporal dependencies of these features, ultimately forming a heart rate feature vector. For blood pressure data, the system focuses not only on the absolute values of systolic and diastolic blood pressure, but also on their ratio and diurnal variation. These features are particularly important for assessing hypertension risk. In practice, when the system detects an abnormal blood pressure pattern (such as a persistent nighttime drop), it promptly reminds the user to consult a doctor and adjust appropriate health recommendations.

[0037] The health profile construction unit 123 is used to construct a health profile based on the processed physiological data, recording the user's health baseline, fluctuation range, and abnormal patterns. This profile contains the user's basic information (age, gender, height, weight, etc.), physiological indicator baseline (resting heart rate, normal blood pressure range, etc.), lifestyle habits (sleep patterns, exercise frequency, etc.), and health risk factors (family medical history, previous illnesses, etc.), providing basic data support for subsequent health assessment and intervention. In family health management, health profiles are a key foundation for personalized intervention. For example, for a 60-year-old hypertensive patient, the system will record their blood pressure baseline (e.g., systolic blood pressure 145mmHg, diastolic blood pressure 95mmHg), fluctuation range (e.g., systolic blood pressure ±15mmHg, diastolic blood pressure ±10mmHg), and sensitivity to environmental factors and lifestyle habits (e.g., sensitivity to sodium intake, sensitivity to room temperature changes, etc.), and adjust the strictness and focus of the intervention plan accordingly.

[0038] like Figure 3 As shown, in a preferred embodiment of the present invention, the multi-objective health intervention subsystem 3 includes an objective function setting module 31 , a constraint condition setting module 32 , an optimization algorithm module 33 and a decision matrix module 34 .

[0039] The objective function setting module 31 is used to set a multi-objective function of nutritional balance, energy balance and medication compliance. The three key objective functions set by this module are: 1. Nutritional balance objective function : Evaluate whether the nutrient distribution in the intervention program is balanced, defined as: , in, Representation scheme Middle The content of nutrients, Indicates the Recommended intake of nutrients, Indicates the The weight of nutrients, Indicates the total amount of nutrients. The target value is set to , that is, the nutritional balance reaches more than 90%. In the family health management scenario, nutritional balance is the basis of a healthy diet. For a type 2 diabetic patient who needs to control blood sugar, the system will set a lower weight for carbohydrate intake (for example, ), while giving higher weight to dietary fiber (e.g. ) to encourage a low-carb, high-fiber diet. At the same time, the system will adjust the recommended intake based on the user's specific situation, for example, adjusting the recommended carbohydrate intake to 75% of that of an average person.

[0040] 2. Energy balance objective function : Assess whether the intervention program balances calorie intake and expenditure, defined as: , in, represents the actual calorie consumption of plan x, Indicates the target calorie consumption. The target value is set to , that is, the error between actual calorie consumption and target calorie consumption is within ±5%. Energy balance is particularly important for weight management. For example, for a user who needs to lose weight, the system will set the daily energy gap to 500kcal (that is, intake is 500kcal less than consumption), so that about 0.5kg can be safely lost per week. At the same time, the system will dynamically adjust the target calorie consumption according to the user's activity level. For example, after the user participates in high-intensity exercise, the recommended energy intake for the day will be automatically increased to ensure adequate nutrition.

[0041] 3. Medication Compliance Objective Function : Assessing user compliance with medication use recommendations, defined as: , in, represents the compliance index of the jth drug in plan x (between 0 and 1), represents the importance weight of the jth drug, and m represents the total number of drugs. The target value is set to , that is, 100% medication compliance. In chronic disease management, medication compliance is directly related to treatment effectiveness. For example, for an elderly user who suffers from both hypertension and diabetes, the weight of antihypertensive drugs (e.g. ) may be higher than those of antidiabetic drugs (e.g. ), as blood pressure control is more critical for preventing cardiovascular and cerebrovascular events. The system will design reasonable medication reminder plans based on the user's lifestyle, such as tying medication take times to meal times or sending reminders when the user habitually checks their phone to improve compliance.

[0042] The constraint setting module 32 is used to set time constraints, health impact constraints, nutritional balance constraints and energy consumption constraints. Specific constraints include: 1. Time constraint: Ensure that the time to complete the intervention task does not exceed the user's available time, expressed as ,in represents the time required for solution x, Represents the user's available time. In family health management, time constraints directly impact the feasibility of intervention plans. For example, for a busy office worker, the system may prioritize short, efficient exercise plans, such as three 10-minute high-intensity interval training sessions per day, over a single 30-minute daily moderate-intensity aerobic exercise session, even though both consume similar amounts of energy.

[0043] 2. Health impact constraint: Ensure that the impact of the intervention plan on user health is within a safe range, expressed as ,in represents the health impact value of plan x, Represents the health impact threshold. Health impact constraints are particularly important for special populations. For example, for patients with heart disease, the system limits the maximum heart rate of the exercise plan to no more than 70% of the maximum safe heart rate for their age (usually calculated as maximum safe heart rate = 220 minus age) to prevent excessive cardiac stress.

[0044] 3. Nutritional balance constraint: Ensure that the nutrient intake in the dietary recommendations does not exceed the safe upper limit, expressed as ,in represents the content of the i-th nutrient in solution x, represents the safe upper limit for the i-th nutrient. In practice, the system will set limits for specific nutrients based on the user's specific circumstances. For example, for patients with renal insufficiency, the upper limit for protein intake is set at 0.8g / kg body weight / day; for patients with hyperuricemia, foods high in purine (such as animal offal and seafood) are restricted.

[0045] 4. Energy consumption constraint: Ensure that the energy consumption in the exercise plan does not exceed the user's maximum tolerance, expressed as ,in represents the energy consumption of solution x, Represents the user's maximum tolerable energy expenditure. This constraint is crucial for ensuring exercise safety. For example, for an obese user starting exercise for the first time, the system will limit maximum energy expenditure to a low level (e.g., 300kcal / day) and gradually increase it as the user's fitness improves to avoid joint injuries and cardiovascular risks.

[0046] The optimization algorithm module 33 is used to perform non-dominated sorting on candidate solutions based on multi-objective functions and constraints, and identify the Pareto optimal solution set. The algorithm flow used in this module is as follows: 1. Generate an initial population P(t) of size 100, with each individual representing a candidate intervention plan. In a family health management system, candidate plans include specific dietary recommendations (e.g., whole-wheat bread, eggs, and milk for breakfast, brown rice, meat, and vegetables for lunch), exercise plans (e.g., brisk walking for 30 minutes daily, strength training twice a week), and medication reminders (e.g., taking antihypertensive medication after breakfast, lipid-lowering medication before bed).

[0047] 2. Perform non-dominated sorting on P(t) to obtain the frontier The principle of non-dominated sorting is: if there is no solution that is better than solution A in all objectives, then solution A is considered non-dominated and belongs to the first frontier. ; Perform the same sorting on the remaining solutions to obtain , and so on.

[0048] 3. Calculate the crowding distance of each individual to maintain population diversity. The crowding distance indicates the density of the surrounding solution space. A larger distance indicates a sparser solution space. Retaining such individuals helps explore a wider solution space.

[0049] 4. Based on user preference vector w=[ , , ]Calculate the weighted objective function F(x x x x), where 、 、 Represents the weights of the three objective functions, which are dynamically adjusted according to the user's historical data. For example, for a hypertensive patient who has poor dietary control but can persist in taking medication, the system may set =0.5 (balanced nutrition), =0.3 (energy balance), =0.2 (medication compliance) to emphasize the importance of balanced nutrition.

[0050] 5. From the Pareto frontier The solution with the largest F(x) is selected as the final solution. This solution balances the three objectives while best meeting the user's personalized needs and preferences.

[0051] The decision matrix module 34 is used to construct a decision matrix for candidate solutions, calculate the comprehensive score through normalization and weighted evaluation, and select the solution with the highest score as the final intervention solution. The decision matrix is constructed using the following method: 1. Construct a decision matrix M, where rows represent different candidate solutions and columns represent different objective function values, for example: , The first column indicates nutritional balance, the second column indicates calorie error, and the third column indicates medication adherence. In this example, the first solution (0.92, 0.03, 1.0) indicates a nutritional balance of 92%, a calorie error of 3%, and 100% medication adherence. The second solution (0.95, 0.06, 0.9) indicates a higher nutritional balance (95%), but a slightly larger calorie error (6%) and lower medication adherence (90%).

[0052] 2. Normalize the decision matrix M to eliminate differences in the dimensions and ranges of the different objective function values, resulting in the normalized matrix N. Normalization uses the maximum-minimum method. For the value aᵢ in the jth column, the normalization formula is: nᵢ = (aᵢ - min(a)) / (max(a) - min(a)). For objectives to be minimized (such as calorie error), the normalization formula is: nᵢ = (max(a) - aᵢ) / (max(a) - min(a)).

[0053] 3. Construct a weighted evaluation matrix W×N, where W=[ , , ] is a weight vector, which is dynamically adjusted based on the user's historical data. For example, if the user is most concerned about nutritional balance, the weight vector may be set to [0.5, 0.3, 0.2]; if the user is most concerned about medication compliance, the weight vector may be set to [0.3, 0.2, 0.5].

[0054] 4. Calculate the overall score for each candidate solution and select the one with the highest score as the final health intervention plan. In family health management, the final selected solution is translated into specific implementation recommendations, such as a detailed weekly meal plan (including specific ingredients and portion sizes), a personalized exercise plan (taking into account user preferences and conditions), and intelligent medication reminders (based on the user's daily routine). The system also dynamically adjusts subsequent intervention plans based on user feedback (such as "I didn't eat breakfast today" or "I didn't have enough time for exercise") to ensure continuity and effectiveness of intervention.

[0055] like Figure 4 As shown, in a preferred embodiment of the present invention, the digital twin verification subsystem 4 includes a user twin model construction module 41, an environment twin model construction module 42, an interaction twin model construction module 43, a time series simulation module 44 and a health status prediction module 45.

[0056] The user twin model construction module 41 is used to construct a virtual user model based on the user's age, gender, weight, underlying diseases, and other attributes. This model uses a parametric modeling approach, describing the user's physiological characteristics and health status through a set of parameters. For example, for a 60-year-old hypertensive patient, the model parameters include: age (60 years old), gender (male), height (170cm), weight (75kg), basal systolic blood pressure (145mmHg), basal diastolic blood pressure (95mmHg), blood sugar level (5.6mmol / L), metabolic rate (1500kcal / day), etc. These parameters are based on the user's actual measurement data and are standardized to construct a virtual user model. In the family health management scenario, the virtual user model is the basis for predicting the effect of interventions. For example, the system can use this model to predict the impact of different interventions on blood pressure: reducing sodium intake may reduce systolic blood pressure by 3-5mmHg, while 30 minutes of aerobic exercise per day may reduce systolic blood pressure by 4-7mmHg. These predictions are based on medical research data and adjusted according to the user's individual characteristics.

[0057] The environmental twin model construction module 42 is used to construct a virtual environment model based on the layout, temperature, humidity, air quality and other parameters of the home environment. The model contains static features of the home environment (such as room layout, furniture position) and dynamic features (such as temperature changes, air quality fluctuations). For example, a typical home environment model includes: living room temperature (23°C), bedroom temperature (22°C), living room humidity (50%), PM2.5 concentration (20μg / m 3 ), noise level (45dB), and other parameters. These parameters are updated in real time by the environmental monitoring module to ensure consistency between the virtual environment model and the actual environment. Environmental models are crucial for predicting the impact of environmental factors on health. For example, for a user with a history of allergies, the system can simulate the impact of various environmental improvement measures (such as using an air purifier or regularly cleaning bed sheets) on indoor allergen levels and predict the extent to which these changes will improve the user's allergic symptoms.

[0058] The interactive twin model construction module 43 is used to construct an interactive model that describes the user's behavior patterns and health responses under different environmental conditions based on the interaction data between the user and the environment. This model uses a conditional probability network to describe the impact of environmental changes on the user's health status. For example, for patients with hypertension, when the indoor temperature rises from 23°C to 28°C, the blood pressure may increase by 5-10 mmHg; when the PM2.5 concentration rises from 20μg / m 3 Increased to 80 μg / m 3 When the user is at home, their heart rate may increase by 5-8 beats per minute. These interaction rules are based on medical research and user historical data and can predict the impact of environmental changes on user health. In home health management, interaction models can guide environmental intervention measures. For example, if the system finds that a user's blood pressure rises significantly when the room temperature is below 18°C, it will recommend keeping the room temperature at least 20°C in winter and remind the user to wear warm clothing before going out to prevent blood pressure fluctuations.

[0059] The time series simulation module 44 simulates a one-week intervention cycle, sampling the user's health indicators at six key time points each day. This module selects these six key time points: 7:00 AM, 10:00 AM, 12:00 PM, 3:00 PM, 6:00 PM, and 9:00 PM before bed. These time points are based on human circadian rhythms and daily routines, covering key daily activities such as waking up, working, eating, and resting. At each time point, the system simulates changes in the user's health indicators, including blood pressure, heart rate, and blood sugar, to form a complete time series of health status. For example, for a patient with type 2 diabetes, the system will pay special attention to post-meal blood sugar changes, adding blood sugar sampling points at 12:00 PM and two hours after 6:00 PM (2:00 PM and 8:00 PM) to assess the impact of different diets on post-meal blood sugar. By analyzing blood sugar fluctuation patterns over the week, the system can adjust the diet (such as increasing dietary fiber and adjusting the carbohydrate distribution) to achieve better blood sugar control.

[0060] The health status prediction module 45 is used to predict the user's health status changes after the intervention period, based on the user's initial health status and intervention behavior. This module uses a state transition model to predict health status changes using the intervention behavior as input. For example, for an overweight user with high blood pressure, if the intervention plan includes sodium restriction and 30 minutes of moderate-intensity exercise daily, the model can predict that the user's weight may decrease by 0.5-1 kg and their blood pressure may decrease by 3-5 mmHg after one week. The predicted results are compared with preset target thresholds, and only if the predicted health improvement indicators exceed the target thresholds is the plan selected as a verified plan. In home health management, health status predictions provide users with an expectation of intervention effects, boosting their confidence and motivation. For example, the system can display the following message to the user: "If you follow the plan, your blood pressure is expected to stabilize within the normal range within two weeks, your weight will decrease by 2-3 kg after one month, and your cardiopulmonary function will significantly improve." This visual prediction is more convincing than abstract health advice and can significantly improve user compliance.

[0061] like Figure 5 As shown, in a preferred embodiment of the present invention, the adaptive interaction subsystem 5 includes a medical terminology recognition module 51 , a dialect adaptation module 52 , a multi-language support module 53 , a personalized push module 54 and an interaction frequency adjustment module 55 .

[0062] The medical term recognition module 51 is used to identify medical terms in user input through medical dictionary matching, context analysis, and model fusion. This module maintains a medical dictionary containing over 500,000 terms, which is used to initially match medical terms in user input. Furthermore, through context analysis, it understands the meaning of medical terms in different contexts and resolves term ambiguity. For example, when a user enters "I have a headache," the system can identify "headache" as a symptom term and, based on the user's historical data and current context, determine the possible cause. In family health management scenarios, medical term recognition is crucial for understanding user needs. For example, when an elderly user describes "I've been feeling a little dizzy and unsteady lately," the system can identify "dizziness" and "unsteady walking" as possible symptoms of vestibular dysfunction or hypotension, providing appropriate advice (such as measuring blood pressure, avoiding sudden standing), or recommending consultation with a doctor. The system also understands different expressions for different user groups. For example, the elderly may use "shortness of breath" to describe breathing difficulties, while children may use "stomachache" to describe various abdominal discomforts.

[0063] The dialect adaptation module 52 is used to extract dialect-specific acoustic and semantic features and convert between dialects and Mandarin. This module is capable of processing voice input from major Chinese dialect regions, including Cantonese, Minnan (Honeymoon), Shanghainese, and Sichuanese. When a user inputs in a dialect, the system first extracts the dialect-specific acoustic features, then converts the input into a standard language form using a dialect-Mandarin mapping model. Finally, semantic understanding and processing are performed. This significantly improves the system's usability across different regions, particularly for elderly users and those who speak different dialects. In family health management, dialect adaptation is particularly important for improving the system's user-friendliness for elderly users. For example, in Fujian, the system can understand health concerns expressed in Minnan dialect, such as "wo tou na hui re" (my head is burning) and "yi na hui mang" (he feels dizzy), and provide corresponding health recommendations. This localized interaction significantly lowers the barrier to entry for technology adoption, enabling more elderly people to benefit from smart health management systems.

[0064] The multilingual support module 53 provides real-time translation capabilities in multiple languages. This module supports input and output in multiple languages, including Chinese, English, Japanese, and Korean, meeting the needs of users from different language backgrounds. When a user inputs in a non-Chinese language, the system first performs language recognition, then translates the input into the standard language (Chinese) used by the system for internal processing. After processing, the input is translated back to the user's original language for output. In diverse family environments, multilingual support significantly expands the system's applicability. For example, in a Chinese-foreign family, if the foreign spouse asks in English, "What should I eat to lower my cholesterol?", the system understands the question, queries the knowledge graph for dietary recommendations for cholesterol-lowering, and responds in English, such as "Eating more oats, beans, nuts, and foods rich in omega-3 can help lower your cholesterol levels." This seamless communication significantly improves the system's user experience.

[0065] The personalized push module 54 is used to optimize push timing, adjust content granularity, and personalize presentation based on user activity patterns, attention patterns, expertise, and preferences. This module analyzes historical user interaction data to construct a user profile, encompassing features such as the user's activity patterns (when they are most active), attention patterns (peak attention periods), expertise (level of medical knowledge), and presentation preferences (preference for brevity or detail). Based on these features, the system optimizes push strategies, for example, delivering concise health reminders to office workers during their lunch break (12:00-1:00 PM) and detailed health advice to retirees between 9:00-10:00 AM. In family health management scenarios, personalized push notifications directly impact the implementation rate of intervention plans. For example, for a busy middle-aged user, the system might deliver a brief reminder of their daily health plan during their commute (e.g., 7:30 AM), dietary recommendations during their lunch break (12:30 PM), and an exercise plan reminder before leaving get off work (5:30 PM). The system dynamically adjusts the content of these reminders based on the user's actual implementation. For a retired elderly person, the system may use more detailed and friendly expressions and push notifications during the hours when the elderly person is usually active (such as 9:00 am and 3:00 pm). At the same time, considering the elderly person's possible memory problems, the system may appropriately increase repeated reminders of important matters.

[0066] The interaction frequency adjustment module 55 monitors user response patterns to push information and dynamically adjusts the push frequency to avoid user fatigue and resistance. This module records user open rates, action rates, and feedback on push information, analyzing user response patterns. A learning algorithm identifies the push frequency that maximizes user engagement. For example, for active users, a push frequency of 3-4 times per day might be used, while for inactive users, this frequency might be reduced to 1-2 times per day. Furthermore, the system sets different push levels based on the importance and urgency of the intervention content. Important and urgent content (such as abnormal blood pressure reminders) is pushed frequently, while general content (such as daily exercise recommendations) is pushed at a regular frequency. In the practical application of home health management systems, an appropriate interaction frequency is crucial for maintaining user engagement. The system learns the optimal frequency for each user. For example, if a user's engagement (open rate and action rate) is highest when pushes are received twice per day, and increases significantly after three pushes, indicating user fatigue, the system will control regular pushes to no more than twice per day, adding additional pushes only in special circumstances (such as when abnormal data is detected). This intelligent frequency adjustment ensures that the system does not burden users while ensuring that important information is delivered in a timely manner.

[0067] In a preferred embodiment of the present invention, the update trigger conditions include: the nutritional balance of the intervention plan reaches more than 90%, the calorie error of the exercise plan is within ±5% and the medication compliance reaches 100%.

[0068] These conditions are based on medical research and user feedback analysis. A nutritional balance of 90% is a high but achievable standard, ensuring comprehensive and balanced nutritional intake to meet the body's daily needs. The calorie tolerance for exercise plans is within ±5%, meaning that the actual calories burned are very close to the target value, ensuring neither excessive fatigue from excessive exercise nor unsatisfactory results from insufficient exercise. 100% medication compliance requires users to take medications exactly as planned, which is particularly important for chronic disease management.

[0069] When these conditions are met, it indicates that the current intervention plan is highly successful, with good user execution and significant results. At this point, the system triggers a dynamic update of the medical knowledge graph, transforming successful intervention experiences into knowledge to guide the generation of future intervention plans. In practical applications of family health management, this dynamic update mechanism enables the system to continuously learn and evolve. For example, the system may discover that for hypertensive patients in a certain region, exercising for 30 minutes daily during the day is more effective than exercising at night; or that millet rice is more effective than white rice in stabilizing postprandial blood sugar levels for blood sugar control. These findings are added to the knowledge graph, improving the effectiveness of future intervention plans. At the same time, the system also updates the knowledge base based on the latest advances in medical research to ensure that health recommendations always align with the latest medical consensus.

[0070] In a preferred embodiment of the present invention, the home smart health management system supports three deployment modes: basic version, standard version and advanced version.

[0071] The Basic Edition implements environmental monitoring, physiological monitoring, and basic health advice. Suitable for users in good health who only need daily health management, this version monitors home environmental parameters and user physiological indicators, providing basic health advice, such as a balanced diet and moderate exercise. The Basic Edition has low hardware requirements and can run on common household devices, making it suitable for large-scale adoption. In actual home use, the Basic Edition can help healthy families establish good living habits and prevent the occurrence of chronic diseases. For example, the system monitors indoor PM2.5 concentrations, reminding users to open windows or use air purifiers when air quality is poor; monitors user activity levels, encouraging sedentary people to get up and move around regularly; and records users' eating habits and provides simple nutritional balance recommendations. While simple, these basic functions are of great value in fostering a healthy lifestyle.

[0072] The Standard Edition builds on the functionality of the Basic Edition by adding a knowledge graph-driven subsystem and a multi-objective health intervention subsystem. This version is suitable for users with specific health needs (such as weight loss and blood pressure control). It can generate personalized intervention plans based on the user's health status and needs, including detailed diet plans, exercise plans, and lifestyle adjustment suggestions. The Standard Edition requires higher computing power and storage space and is suitable for families with high requirements for health management. In the family health management scenario, the Standard Edition can meet the needs of most users with chronic health problems. For example, for a patient with mild hypertension, the system will combine their blood pressure data, lifestyle habits, and dietary preferences to generate a personalized DASH diet plan (a low-sodium, high-potassium, vegetable-rich diet) and a suitable exercise plan, and dynamically adjust the plan based on the user's implementation status and blood pressure changes. The multi-objective optimization function of the Standard Edition ensures that the intervention plan strikes a balance between effectiveness and feasibility, greatly improving user compliance and health outcomes.

[0073] The Advanced Edition builds on the functionality of the Standard Edition by adding a digital twin verification subsystem and multimodal medical terminology recognition. Suitable for patients with chronic conditions or those at higher health risks, this edition provides predictive verification of intervention plans to ensure their effectiveness, while also offering a more natural and intelligent interactive experience. The Advanced Edition requires powerful computing resources and can be deployed locally or in the cloud, tailored to the user's privacy needs and computing resources. The Advanced Edition's advantages are particularly evident in complex home health management scenarios. For example, for an elderly patient with both type 2 diabetes and coronary artery disease, the system builds a precise digital twin model to simulate the combined effects of different interventions on blood sugar, blood lipids, and cardiovascular health, identifying the optimal balance. Furthermore, the Advanced Edition's multimodal medical terminology recognition feature enables elderly patients to describe their symptoms in natural language (e.g., "I've been experiencing numbness in my feet lately"), which the system then understands may be an early symptom of diabetic peripheral neuropathy and provides appropriate recommendations or a recommendation for medical attention. These advanced predictive and interactive capabilities significantly improve the accuracy of health management and the user experience.

[0074] like Figure 6 As shown, the family intelligent health management method provided by the present invention includes the following steps: Step S1: Collect home environment parameters and user physiological indicators. Specifically, the PM2.5 sensor and temperature sensor collect environmental data every 30 seconds; the user's heart rate, blood pressure, blood oxygen, body temperature and other physiological indicators, as well as sleep, activity, stress and other behavioral data are collected through smart bracelets and smartphones. In the daily application of family health management, this step ensures that the system can fully understand the user's health status and living environment. For example, when the system detects that the PM2.5 concentration in the user's living environment continues to exceed 50μg / m 3At the same time, the user's heart rate increases by 5-10 beats / minute compared to usual, which may indicate that environmental pollution has affected the user's cardiopulmonary function. The system will remind the user to open windows for ventilation or reduce outdoor activities.

[0075] Step S2: Based on environmental parameters and user physiological indicators, a three-layer medical knowledge graph is constructed, consisting of an entity layer, a relationship layer, and a knowledge layer. Specifically, entity recognition technology is used to convert entity names in medical literature into standard medical identifiers; a medical knowledge graph consisting of an entity layer, a relationship layer, and a knowledge layer is established; and based on the user's health status and needs, relevant medical knowledge is retrieved to support the generation of intervention plans. In practical applications, knowledge graphs are the basis for personalized intervention. For example, when the system detects that the user's blood pressure is elevated, all nodes related to hypertension are retrieved from the knowledge graph, including possible causes (such as a high-sodium diet, lack of exercise, and high stress), potential risks (such as cardiovascular and cerebrovascular events), and effective intervention measures (such as the DASH diet, aerobic exercise, and stress management). The most appropriate intervention strategy is then selected based on the user's specific situation.

[0076] Step S3: Based on the medical knowledge graph, a set of candidate intervention plans is generated by setting a multi-objective function consisting of nutritional balance, energy balance, and medication compliance goals. Specifically, a nutritional balance goal (above 90% nutritional balance), an energy balance goal (the error between actual calorie consumption and target calorie consumption is within ±5%), and a medication compliance goal (100% medication compliance) are set. Time constraints, health impact constraints, nutritional balance constraints, and energy consumption constraints are set. The candidate plans are non-dominated sorted to identify the Pareto optimal solution set. A decision matrix for the candidate plans is constructed, and a comprehensive score is calculated through normalization and weighted evaluation. The plan with the highest score is selected as the final intervention plan. In family health management, multi-objective optimization ensures the comprehensive effectiveness of intervention plans. For example, for a middle-aged woman with mild hypertension who needs to lose weight, the system must consider not only the weight loss goal (calorie intake control) but also blood pressure control (low-sodium diet), as well as the user's personal preferences and time constraints. Through multi-objective optimization, the system is able to generate a plan that balances various needs, such as recommending low-calorie, low-sodium but good-tasting recipes, and exercise plans that combine user interests (such as dancing) and time arrangements.

[0077] Step S4: Establish user twin models, environment twin models, and interaction twin models, perform predictive validation on the candidate intervention scenarios, and select those with predicted health improvement indicators above the target threshold as the validated scenarios. Specifically, a virtual user model is constructed based on user attributes such as age, gender, weight, and underlying medical conditions; a virtual environment model is constructed based on parameters such as the home environment's layout, temperature, humidity, and air quality; and an interaction model is constructed based on user-environment interaction data. A one-week intervention cycle is simulated, with user health indicators sampled at six key time points each day. Based on the user's initial health status and intervention behavior, changes in the user's health status after the intervention cycle are predicted. Digital twin validation is a major innovation of this invention, making intervention effects predictable and significantly improving the safety and effectiveness of the intervention. For example, for a diabetic patient, the system can simulate the effects of different diets on their blood sugar, comparing whole-wheat bread with regular bread, and brown rice with white rice, to identify the most suitable dietary structure for the user. This predictive validation avoids the risk of blood sugar fluctuations that can occur during trial and error, while providing users with clear expectations and increasing motivation to act.

[0078] Step S5: Based on multimodal recognition of medical terminology and user profile analysis, the verified solution is pushed to the user through a personalized push strategy. Specifically, medical terminology in user input is identified through medical dictionary matching, context analysis, and model fusion. The unique acoustic and semantic features of dialects are extracted to enable conversion between dialects and Mandarin. Real-time translation is provided for multiple languages. Push notifications are optimized based on user activity patterns, attention patterns, expertise, and preferences, adjusting content granularity and personalized presentation. The user's response patterns to push notifications are monitored to dynamically adjust push frequency. Adaptive interaction significantly improves the system's user experience and accessibility. For example, for an elderly user who speaks a predominantly dialect-based health issue, the system can understand the dialect and respond using familiar expressions, such as "Your systolic blood pressure is high. Limiting sodium intake is recommended" to "Your blood pressure is a bit high. Reduce salt intake and eat less pickled vegetables. This is good for your health." Furthermore, the system selects push notifications during the hours when the elderly are most receptive (e.g., 8:00-9:00 AM) based on their sleep patterns and attention patterns, using larger fonts and simpler language.

[0079] Step S6: Collect user feedback and, when preset update trigger conditions are met, trigger a dynamic update of the medical knowledge graph. Specifically, user feedback on the implementation and health effects of the intervention plan is collected. A knowledge graph update is triggered when the nutritional balance of the intervention plan reaches above 90%, the calorie tolerance of the exercise plan is within ±5%, and medication compliance reaches 100%. Entity, relationship, attribute, and data updates are performed to ensure the timeliness and accuracy of medical knowledge. Closed-loop feedback is key to continuous system optimization. In the long-term application of family health management, the system will accumulate a large amount of user feedback data, from which it learns patterns and regularities in intervention effects. For example, the system may discover that morning exercise on an empty stomach is more effective for weight loss than exercise after meals among users of a certain age group; or that certain food combinations (such as a specific ratio of protein to dietary fiber) are particularly effective for blood sugar control. These findings are incorporated into the system through the knowledge graph's dynamic update mechanism, continuously improving the accuracy and personalization of intervention plans.

[0080] The above embodiments are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. Home intelligent health management system, characterized by: include: The data acquisition subsystem includes an environmental monitoring module and a physiological monitoring module, which are used to collect home environment parameters and user physiological indicators; A knowledge graph driving subsystem, connected to the data acquisition subsystem, is configured to receive the environmental parameters and the user's physiological indicators, construct a three-layer medical knowledge graph comprising an entity layer, a relationship layer, and a knowledge layer, and dynamically update the medical knowledge graph based on preset update trigger conditions; A multi-objective health intervention subsystem, connected to the knowledge graph driving subsystem, is used to generate a set of candidate intervention plans based on the medical knowledge graph by setting a multi-objective function of nutritional balance goals, energy balance goals, and medication compliance goals; A digital twin verification subsystem, connected to the multi-objective health intervention subsystem, is used to establish a user twin model, an environment twin model, and an interaction twin model, perform predictive verification on the candidate intervention scheme set, and select the scheme with the predicted health improvement index higher than the target threshold as the verified scheme; An adaptive interaction subsystem, connected to the digital twin verification subsystem, is configured to receive the verification plan, push the verification plan to the user through a personalized push strategy based on multimodal recognition of medical terms and user profile analysis, and collect user feedback information; Among them, the digital twin verification subsystem is also connected to the knowledge graph driving subsystem to transmit the user feedback information to the knowledge graph driving subsystem to trigger the dynamic update of the medical knowledge graph.

2. The home intelligent health management system according to claim 1, characterized in that: The knowledge graph driven subsystem includes: Entity recognition module, which converts entity names in medical literature into standard medical identifiers; A three-layer structure building module is used to build a medical knowledge graph consisting of an entity layer, a relationship layer, and a knowledge layer. The entity layer includes drugs, symptoms, diseases, departments, equipment, ingredients, acupoints, and action entities; the relationship layer includes attribute relationships and inter-entity relationships; and the knowledge layer includes diagnostic rules, treatment plans, and health recommendations. Adaptive update module, used to perform entity update, relationship update, attribute update and data update when the preset update trigger conditions are met; The knowledge query module is used to retrieve relevant medical knowledge based on the user's health status and needs, and support the generation of intervention plans.

3. The home intelligent health management system according to claim 1, characterized in that: The environmental monitoring module includes: High-frequency sampling unit, used to collect home environment data every 30 seconds through PM2.5 sensor and temperature sensor; A convolution feature extraction unit, configured to perform convolution and pooling processing on the environmental data to extract local environmental features; Environmental grading assessment unit, used to compare the extracted environmental characteristics with preset health standards and classify the environmental status into three levels: low risk, medium risk and high risk; The early warning trigger unit is used to generate early warning information when the environmental parameters exceed the preset threshold value and incorporate the early warning information into the intervention plan.

4. The home intelligent health management system according to claim 1, characterized in that: The physiological monitoring module includes: Multi-source data acquisition unit, used to collect users' physiological indicators such as heart rate, blood pressure, blood oxygen, body temperature, as well as behavioral data such as sleep, activity, and stress through smart bracelets and smartphones; a time series feature extraction unit, configured to process the physiological indicators and the behavioral data through a time series feature extraction network to capture time dependencies and important features; The health record construction unit is used to construct a health record that records the user's health baseline, fluctuation range and abnormal pattern based on the processed physiological data.

5. The home intelligent health management system according to claim 1, characterized in that: The multi-target health intervention subsystem includes: Objective function setting module, used to set multi-objective functions including nutritional balance target, energy balance target and medication compliance target; Constraint setting module, used to set time constraints, health impact constraints, nutritional balance constraints and energy consumption constraints; An optimization algorithm module, configured to perform non-dominated sorting on candidate solutions based on the multi-objective function and the constraints, and identify a Pareto optimal solution set; The decision matrix module is used to construct the decision matrix of candidate solutions, calculate the comprehensive score through normalization processing and weighted evaluation, and select the solution with the highest score as the final intervention solution.

6. The home intelligent health management system according to claim 1, characterized in that: The digital twin verification subsystem includes: User twin model construction module, used to build a virtual user model based on user attributes such as age, gender, weight, and underlying diseases; The environmental twin model construction module is used to build a virtual environment model based on the layout, temperature, humidity, air quality and other parameters of the home environment; The interactive twin model construction module is used to build an interactive model that describes the user's behavior patterns and health responses under different environmental conditions based on the interaction data between the user and the environment; The time series simulation module is used to simulate a one-week intervention cycle and sample user health indicators at six key time points every day; The health status prediction module is used to predict the changes in the user's health status after the intervention cycle based on the user's initial health status and intervention behavior.

7. The home intelligent health management system according to claim 1, characterized in that: The adaptive interaction subsystem includes: Medical term recognition module, which is used to identify medical terms in user input through medical dictionary matching, context analysis, and model fusion; The dialect adaptation module is used to extract the unique acoustic and semantic features of the dialect and realize the conversion between the dialect and Mandarin; Multi-language support module, used to provide real-time translation function in multiple languages; A personalized push module, which optimizes push timing, adjusts content granularity, and personalizes presentation based on user activity patterns, attention patterns, expertise, and preferences. The interaction frequency adjustment module is used to monitor the user's response pattern to push information and dynamically adjust the push frequency to avoid user fatigue and resistance.

8. The home intelligent health management system according to claim 1, characterized in that: The update trigger conditions include: the nutritional balance of the intervention plan reaches more than 90%, the calorie error of the exercise plan is within plus or minus 5%, and the medication compliance reaches 100%.

9. The home intelligent health management system according to claim 1, characterized in that: The home smart health management system supports three deployment modes: basic version, standard version and advanced version, among which: The basic version implements environmental monitoring, physiological monitoring and basic health advice functions; The standard version adds a knowledge graph driven subsystem and a multi-target health intervention subsystem based on the basic version. The advanced version adds a digital twin verification subsystem and multimodal recognition function of medical terms based on the functions of the standard version.

10. A family intelligent health management method, characterized in that: include: Collect home environment parameters and user physiological indicators; Based on the environmental parameters and the user's physiological indicators, a three-layer medical knowledge graph is constructed, comprising an entity layer, a relationship layer, and a knowledge layer; Based on the medical knowledge graph, a set of candidate intervention plans is generated by setting a multi-objective function of nutritional balance goals, energy balance goals, and medication compliance goals; Establish user twin models, environment twin models, and interaction twin models, perform predictive verification on the candidate intervention plan set, and select plans with predicted health improvement indicators above the target threshold as verified plans; Based on multimodal recognition of medical terms and user portrait analysis, the verification solution is pushed to the user through a personalized push strategy; Collect user feedback information and trigger the dynamic update of the medical knowledge graph when the preset update trigger conditions are met; Among them, the update trigger conditions include: the nutritional balance of the intervention plan reaches more than 90%, the calorie error of the exercise plan is within the range of plus or minus 5%, and the medication compliance reaches 100%.

Images