Health management scheme recommendation system based on big data analysis

Abstract

The invention relates to the technical field of health management systems, and discloses a health management scheme recommendation system based on big data analysis. The system comprises a health data acquisition module, a health characteristic quantification module, a health state identification module, a health scheme prediction module and a health parameter coupling module. The health data acquisition module synchronously acquires three types of time sequence data of physiological indexes, health behaviors and environmental exposure of a user and performs timestamp alignment; a health feature quantification module extracts features from the aligned data and generates corresponding feature matrixes and vectors; the health state recognition module classifies the feature data according to a preset rule and generates a health state label set; the health scheme prediction module is combined with a health intervention measure knowledge base to generate an initial scheme set through an association rule mining algorithm; and the health parameter coupling module corrects the intervention intensity parameter of the initial scheme based on the mapping relationship between the human physiological response and the behavior intervention parameter. According to the system, multi-dimensional health data can be integrated, and an accurate personalized health management scheme is generated.

Description

Technical Field

[0001] This invention relates to the field of health management system technology, specifically a health management plan recommendation system based on big data analysis. Background Technology

[0002] With the increasing health awareness of residents and the widespread use of mobile medical devices, health management is gradually shifting from traditional post-event intervention to pre-event prevention, and users' demand for personalized health management solutions is growing. Currently, most health management products on the market can only collect and display single types of health data, such as physiological indicators like heart rate and steps, or simply recording health behavior data like diet and sleep. They cannot simultaneously integrate multi-dimensional data such as physiological indicators, health behaviors, and environmental exposures, resulting in data that is too singular to comprehensively reflect a user's true health status.

[0003] Even though some products attempt to collect multi-dimensional health data, they fail to effectively align the timestamps of data from different sources and of different types. Since physiological indicator data is typically collected in minutes or seconds, health behavior data is often recorded in hours or days, and environmental exposure data (such as air quality, temperature, and humidity) is collected at different frequencies, this inconsistency in the time dimension of the data makes it impossible to accurately analyze the correlations between the various dimensions of data, thus affecting the accuracy of health status assessments.

[0004] In the health status identification process, most existing technologies rely on simple threshold judgment methods, such as using whether blood pressure exceeds the normal range as the sole criterion for judging cardiovascular health status, without considering the interaction between physiological indicators, health behaviors, and environmental exposures. For example, when a user engages in high-intensity exercise in a high-temperature environment, an increase in heart rate may be a normal physiological response, but if judged solely based on heart rate thresholds, it may be misjudged as an abnormal heart health, leading to inaccurate health status label generation.

[0005] In terms of generating health management plans, existing systems mostly adopt a fixed template recommendation approach, providing the same intervention plan for the same type of health problem without adjusting for individual differences in users' physiological responses. For example, even with weight management, different users have significantly different physiological adaptability to dietary control and exercise intensity. A fixed plan may be too intense, making it difficult for users to adhere to, or too weak, failing to achieve the desired effect. Furthermore, existing systems lack effective utilization of health intervention knowledge bases and cannot use data mining to discover potential correlations between health status and intervention measures. This results in recommendations that are insufficiently targeted and effective, failing to meet users' personalized and precise health management needs. Summary of the Invention

[0006] The purpose of this invention is to provide a health management solution recommendation system based on big data analysis to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides a health management solution recommendation system based on big data analysis, the system comprising: The health data acquisition module is used to synchronously collect users' physiological indicator time-series data, health behavior time-series data, and environmental exposure time-series data, and to align the physiological indicator time-series data, health behavior time-series data, and environmental exposure time-series data with timestamps. The health feature quantification module is used to extract features from the time-series data of physiological indicators, health behavior, and environmental exposure output by the health data acquisition module after timestamp alignment, and generate physiological indicator feature matrix, health behavior quantification matrix, and environmental exposure feature vector, respectively. The health status identification module is used to classify the physiological indicator feature matrix, health behavior quantification matrix and environmental exposure feature vector output by the health feature quantification module based on the preset health status classification rules, and generate a set of user health status tags. The health plan prediction module is used to predict and generate an initial set of health management plans based on the set of user health status tags output by the health status recognition module, combined with a pre-stored health intervention knowledge base, and through an association rule mining algorithm. The health parameter coupling module is used to perform physiological adaptation correction on the intervention intensity parameters in the initial health management plan set output by the health plan prediction module, based on the mapping relationship between human physiological response parameters and behavioral intervention parameters.

[0008] Preferably, the health data acquisition module includes: A multi-source data synchronization unit is used to synchronously collect the time-series data of physiological indicators, health behavior, and environmental exposure. The time-series data of physiological indicators includes heart rate variability and blood glucose concentration. The time-series data of health behavior includes exercise intensity and sleep quality. The time-series data of environmental exposure includes air quality index and temperature and humidity change. The timestamp alignment unit is used to perform timestamp alignment on the physiological indicator time-series data, health behavior time-series data, and environmental exposure time-series data output by the multi-source data synchronization unit using an interpolation compensation method. The data quality cleaning unit is used to remove outlier data points from the time-series data of physiological indicators, health behaviors, and environmental exposures output by the timestamp alignment unit, and generate a standardized health time-series dataset.

[0009] Preferably, the health characteristic quantification module includes: The physiological indicator analysis unit is used to extract time-domain variability features and frequency-domain energy distribution features from the time-series data of physiological indicators in the standardized health time-series dataset, and generate the physiological indicator feature matrix. The behavior pattern quantification unit is used to extract the distribution features of behavior duration and the gradient features of intensity change from the health behavior time series data in the standardized health time series dataset, and generate the health behavior quantification matrix. The environmental exposure parsing unit is used to extract exposure frequency features and cumulative exposure amount features from the environmental exposure time series data in the standardized health time series dataset, and generate the environmental exposure feature vector.

[0010] Preferably, the health status recognition module includes: The status classification unit is used to classify the physiological indicator feature matrix, health behavior quantification matrix and environmental exposure feature vector according to a preset metabolic syndrome risk level threshold, and generate a preliminary health status label. The tag fusion unit is used to analyze the correlation between tags based on the preliminary health status tags output by the state classification unit, and fuse them to generate the user health status tag set.

[0011] Preferably, the health plan prediction module includes: The knowledge base matching unit is used to retrieve matching basic intervention item entries from the health intervention measure knowledge base based on the user health status tag set. The association rule mining unit is used to discover the combination patterns of measures based on the basic intervention measure items output by the knowledge base matching unit, and generate the initial health management plan set by using a frequent itemset mining algorithm.

[0012] Preferably, the health parameter coupling module includes: The physiological response mapping unit is used to adaptively modify the exercise intensity parameters in the initial health management program set based on the heart rate recovery coefficient in the mapping relationship between the human physiological response parameters and behavioral intervention parameters. The behavioral tolerance correction unit is used to perform tolerance correction on the sleep sensitivity coefficient in the mapping relationship between the human physiological response parameters and behavioral intervention parameters, based on the sleep sensitivity coefficient.

[0013] Preferably, the system further includes: a health plan optimization module, used to adjust the initial health management plan set based on the corrected intervention intensity parameters output by the health parameter coupling module, combined with the user's historical plan compliance records, using a feature space mapping method, to generate an optimized health management plan set; the health plan optimization module includes: The feature space construction unit is used to construct a multi-dimensional feature space by combining the modified intervention intensity parameter with the execution completion rate index in the user's historical plan compliance record. The mapping adjustment unit is used to identify the optimization direction in the multidimensional feature space generated by the feature space construction unit through the nearest neighbor propagation clustering algorithm, and adjust the initial health management scheme set to generate the optimized health management scheme set.

[0014] Preferably, the system further includes: a health pathway generation module, used to optimize the health management plan set output by the health plan optimization module, and combine it with the user's daily behavior pattern characteristics to optimize the temporal arrangement and combination of intervention measures, thereby generating a health management execution pathway plan; the health pathway generation module includes: The time-series pattern analysis unit is used to extract the idle time distribution features and behavior transition nodes from the user's schedule behavior pattern features; The pathway optimization unit is used to generate the health management execution pathway scheme based on the idle time period distribution characteristics and behavior transition nodes output by the time series pattern analysis unit, combined with the intervention measures in the optimized health management scheme set, through a constraint satisfaction algorithm.

[0015] Preferably, the system further includes: a health dynamic update module, used to receive the latest user feedback data in real time, and update the user's health status tag set based on the intervention nodes in the health management execution path plan output by the health path generation module through an incremental feature learning method; the health dynamic update module includes: The feedback feature extraction unit is used to extract physiological indicator shift features and behavioral compliance change features from the latest user feedback data. An incremental learning unit is used to incrementally update the user's health status label set at the intervention node of the health management execution pathway scheme based on the physiological index shift features and behavioral compliance change features output by the feedback feature extraction unit.

[0016] Preferably, the system further includes: a health decision output module, used to generate a final health management recommendation instruction based on the updated user health status tag set output by the health dynamic update module and the health management execution path scheme output by the health path generation module; the health decision output module includes: The scheme integration unit is used to perform feature association and alignment between the updated user health status tag set and the health management execution path scheme. The instruction generation unit is used to generate the final health management recommendation instruction, which includes a sequence of intervention measures and execution time nodes, based on the correlation alignment results output by the scheme integration unit.

[0017] Compared with the prior art, the beneficial effects of the present invention are: By synchronously collecting users' physiological indicators, health behaviors, and environmental exposure time-series data through the health data acquisition module and aligning them with timestamps, the limitations of traditional health management products in terms of single data collection dimensions and inconsistent time dimensions are overcome. The synchronous collection and timestamp alignment of multi-dimensional data comprehensively captures the correlation information between users' physiological states, behavioral habits, and external environments at different time points, providing a complete data foundation for accurate analysis of users' health status and avoiding misjudgments of health status due to incomplete data or time misalignment.

[0018] The health feature quantification module extracts features from the timestamp-aligned multi-dimensional data, generating physiological indicator feature matrices, health behavior quantification matrices, and environmental exposure feature vectors. This transforms the originally scattered and disordered raw data into structured and analyzable feature data. This feature quantification process effectively mines key information from various dimensions of data. For example, it extracts deep features such as heart rate variability and blood pressure fluctuation trends from physiological indicator time-series data; quantifies behavioral features such as the rationality of dietary structure and regularity of exercise from health behavior time-series data; and extracts key environmental factors affecting health from environmental exposure data. This allows subsequent health status identification processes to make judgments based on more accurate and valuable data features, improving the scientific rigor of health status identification.

[0019] The health status recognition module, based on preset health status classification rules, categorizes multi-dimensional feature data and generates a set of user health status tags, changing the traditional approach of relying on a single threshold to determine health status. This module fully considers the interaction between physiological indicators, health behaviors, and environmental exposures. For example, when assessing a user's sleep quality, it not only considers behavioral data such as sleep duration and sleep stages but also combines physiological indicators such as nighttime heart rate and respiratory rate with environmental exposure data such as bedroom temperature and noise. This comprehensive approach generates sleep quality-related health status tags, enabling the tags to more comprehensively and accurately reflect the user's true health status and providing a reliable basis for the generation of personalized health plans.

[0020] The health plan prediction module, combined with a pre-stored knowledge base of health intervention measures, uses an association rule mining algorithm to generate an initial set of health management plans based on a set of health status labels, breaking away from the traditional model of recommending plans using fixed templates. The association rule mining algorithm can uncover potential associations between health statuses and intervention measures from the knowledge base. For example, it discovers a strong association between "high-salt diet + hypertension" and "low-sodium diet intervention + light aerobic exercise," thus recommending more targeted initial plans for users with different combinations of health status labels, avoiding plan homogenization and meeting the diverse health management needs of users.

[0021] The health parameter coupling module, based on the mapping relationship between human physiological response parameters and behavioral intervention parameters, performs physiological adaptation adjustments to the intervention intensity parameters in the initial health management plan, further enhancing the personalization and feasibility of the plan. Different users exhibit different physiological responses to the same intervention; for example, with the same exercise intensity, younger users may experience no significant discomfort, while middle-aged and elderly users may experience excessively high heart rates. This module, by adjusting the intervention intensity parameters, allows the health management plan to better adapt to the individual physiological characteristics of users. This avoids health risks caused by inappropriate intervention intensity, improves user adherence to the plan, and helps users implement the health management plan long-term, thereby achieving better health management results. Attached Figure Description

[0022] Figure 1 This is a sequence diagram of the health management solution recommendation system based on big data analysis described in this invention; Figure 2 A flowchart illustrating the operation of the health metric quantification module; Figure 3 A flowchart illustrating how the health status recognition module works; Figure 4 A flowchart illustrating the workflow of the health plan optimization module; Figure 5 A flowchart illustrating the workflow of the health dynamics update module. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Please see Figure 1 This invention provides a health management solution recommendation system based on big data analysis, the system comprising: By collecting multi-source health data, extracting features, identifying states, predicting solutions, and coupling parameters, the system generates and optimizes personalized health management plans. First, the health data acquisition module synchronously collects time-series data on users' physiological indicators, health behaviors, and environmental exposures, and aligns these data with timestamps. The health feature quantification module extracts features from the aligned data, generating a physiological indicator feature matrix, a health behavior quantification matrix, and an environmental exposure feature vector. The health state identification module classifies these features based on preset health state classification rules, generating a set of user health state labels. The health solution prediction module, combined with a pre-stored health intervention knowledge base, uses an association rule mining algorithm to predict and generate an initial set of health management plans. The health parameter coupling module, based on the mapping relationship between human physiological response parameters and behavioral intervention parameters, physiologically adapts the intervention intensity parameters in the initial health management plan set, ultimately outputting a health management plan that conforms to the user's physiological characteristics.

[0025] Example 1: See Figure 2 The system collects time-series data on users' physiological indicators, health behaviors, and environmental exposures through a multi-source data synchronization unit. Physiological indicator time-series data includes heart rate variability and blood glucose concentration sequences, collected via wearable devices such as smart bracelets and continuous glucose monitors, recording data points every minute. Health behavior time-series data includes exercise intensity and sleep quality sequences. Exercise intensity data is collected via accelerometers, using triaxial acceleration values ​​to represent exercise intensity. Sleep quality data is collected via motion recorders, recording sleep stage transitions and durations. Environmental exposure time-series data includes air quality index and temperature and humidity change sequences, collected via an environmental sensor network. The air quality index includes PM2.5 and PM10 particulate matter concentrations, and the temperature and humidity data records minute-by-minute changes in ambient temperature and humidity.

[0026] The timestamp alignment unit uses an interpolation compensation method to process the data output by the multi-source data synchronization unit. Due to differences in sampling frequencies and start times among different sensors, the timestamps of the original data may have offsets ranging from milliseconds to minutes. The interpolation compensation method employs a linear interpolation algorithm, using the data stream with the highest sampling frequency as the reference time axis. Compensation values ​​for other data streams are calculated proportionally based on the timestamp differences, aligning all data points on a unified time axis. For example, for missing heart rate data within a certain second, the average value of the values ​​from the two seconds before and after is used to fill the gap; for temperature and humidity data collected every minute, the values ​​are distributed to each second within that minute to maintain data continuity.

[0027] The data quality cleaning unit uses an outlier detection algorithm to process the timestamp-aligned data. Outlier detection is based on the interquartile range (IQR) method in statistics, calculating the upper and lower quartiles of each data sequence to define a reasonable data range. Data points outside this range are marked as outliers and removed. For example, in a heart rate variability series, the normal heart rate range is set to 40 to 180 beats per minute; data points outside this range are considered outliers. In a blood glucose concentration series, the normal range is set to 3.9 to 11.1 mmol / L; values ​​outside this range are removed. After removing outliers, the data quality cleaning unit uses linear interpolation to compensate for missing data, generating a standardized health time-series dataset.

[0028] The physiological indicator analysis unit in the health feature quantification module extracts features from the time-series data of physiological indicators in the standardized health time-series dataset. Time-domain variability features include statistics such as mean, standard deviation, and root mean square difference, used to characterize the fluctuation characteristics of physiological indicators. For example, calculating the standard deviation of adjacent heartbeat intervals for a heart rate variability sequence reflects the regulatory function of the autonomic nervous system; calculating the standard deviation of intraday blood glucose fluctuations for a blood glucose concentration sequence characterizes blood glucose stability. Frequency-domain energy distribution features convert the time-domain signal to a frequency-domain signal using a fast Fourier transform, extracting the energy ratio of low-frequency to high-frequency components. For example, performing spectral analysis on a heart rate variability signal calculates the ratio of low-frequency power to high-frequency power, reflecting the balance between the sympathetic and parasympathetic nervous systems. These time-domain and frequency-domain features together constitute the physiological indicator feature matrix.

[0029] The behavioral pattern quantification unit extracts features from time-series data of health behaviors. Behavioral duration distribution features include the cumulative time distribution of exercise at different intensities and the proportion of duration in each sleep stage. For example, it analyzes the daily cumulative time of low-intensity, medium-intensity, and high-intensity exercise in exercise intensity sequences; and it analyzes the duration proportions of light sleep, deep sleep, and REM sleep in sleep quality sequences. The intensity change gradient feature characterizes the dynamic characteristics of behavioral patterns by calculating the rate of change in behavioral intensity. For example, it calculates the intensity change gradient per unit time in exercise intensity sequences to reflect fluctuations in exercise intensity; and it calculates the frequency of sleep stage transitions in sleep quality sequences to characterize sleep stability. These features constitute the health behavior quantification matrix.

[0030] The environmental exposure analysis unit extracts exposure frequency and cumulative exposure characteristics from environmental exposure time-series data. Exposure frequency characteristics are characterized by statistically analyzing the number of exposure events occurring above a specific threshold. For example, for an air quality index series, it counts the number of times PM2.5 concentrations exceed 75 micrograms per cubic meter; for a temperature and humidity change series, it counts the number of times temperatures exceed 30 degrees Celsius or humidity falls below 30%. Cumulative exposure characteristics characterize the cumulative effect of environmental impacts by calculating the total exposure over a certain period. For example, it calculates the cumulative daily exposure to PM2.5 concentrations; it calculates the cumulative time for temperatures to deviate from the comfortable range (20-26 degrees Celsius). These characteristics constitute the environmental exposure feature vector.

[0031] Example 2: See Figure 3 The health status identification module's status classification unit classifies input features based on preset metabolic syndrome risk level thresholds. These thresholds are set according to international clinical guidelines and include multiple dimensions of criteria. For example, the blood pressure dimension sets systolic blood pressure of 130 mmHg and diastolic blood pressure of 85 mmHg as thresholds; the blood glucose dimension sets fasting blood glucose of 5.6 mmol / L as a threshold; and the blood lipid dimension sets high-density lipoprotein cholesterol of 1.0 mmol / L for men and 1.3 mmol / L for women as thresholds. The status classification unit compares the blood pressure feature value in the physiological indicator feature matrix with the corresponding threshold. When the systolic blood pressure feature value consistently exceeds 130 mmHg, it is marked as an abnormal blood pressure state. Similarly, the blood glucose feature value is compared with the 5.6 mmol / L threshold, and the blood lipid feature value is compared with the corresponding threshold, generating preliminary health status labels. These labels include, but are not limited to, discrete status identifiers such as "high blood pressure," "borderline blood glucose," and "abnormal blood lipids."

[0032] The label fusion unit analyzes the correlation between these preliminary health status labels using a state transition probability model. This model, trained on a large amount of historical health data, captures the transition patterns between different health states. For example, the model might reveal a strong correlation between "high blood pressure" and "borderline blood sugar," indicating a higher probability of developing metabolic syndrome when both states occur simultaneously. The label fusion unit calculates the joint probability between these states, fusing the relevant preliminary labels into a more comprehensive health status description, such as merging "high blood pressure" and "borderline blood sugar" into "pre-metabolic syndrome risk." This process generates a set of user health status labels containing multiple levels of status descriptions, ranging from specific physiological abnormalities to comprehensive health risk states.

[0033] The knowledge base matching unit in the health plan prediction module searches and matches data from a pre-stored health intervention knowledge base based on the user's set of health status tags. This knowledge base contains thousands of clinically validated intervention entries, each labeled with an applicable health status tag. For example, for the "high blood pressure" status, the knowledge base might include measures such as "aerobic exercise intervention" and "low-sodium diet recommendations"; for the "borderline blood sugar" status, it might include measures such as "carbohydrate control" and "regular blood sugar monitoring." The knowledge base matching unit calculates tag similarity to retrieve the basic intervention entry from the knowledge base that best matches the user's current health status tag.

[0034] The association rule mining unit, based on these matching basic intervention items, employs a frequent itemset mining algorithm to discover patterns in the combination of interventions. The algorithm analyzes the frequency of different interventions occurring simultaneously in historical success cases, identifying frequently used combinations. For example, the algorithm might find that "aerobic exercise intervention" and "low-sodium diet recommendations" frequently appear together in programs targeting blood pressure issues, and "carbohydrate control" and "regular blood glucose monitoring" are often used together for blood glucose management. Through the mining of these frequent itemsets, the system generates an initial set of health management programs, each containing a set of complementary interventions. These programs consider not only single health problems but also the correlations between multiple health status labels, forming comprehensive health management recommendations.

[0035] The system accurately identifies users' health status characteristics through multi-level status classification and label fusion. Then, based on a knowledge base and association rule mining, it generates combinations of intervention measures with inherent logical connections. This method can filter out validated and effective solutions from a large number of possible combinations, providing users with personalized health management recommendations. Through this data-driven approach, the system standardizes and automates the complex process of health status identification and solution generation, providing reliable technical support for health management.

[0036] Example 3: See Figure 4The system receives an initial set of health management plans from the health plan prediction module, which contains multiple combinations of interventions targeting the user's health status tags. The physiological response mapping unit in the health parameter coupling module adaptively modifies the exercise intensity parameters in the initial plans based on a predefined mapping relationship between human physiological response parameters and behavioral intervention parameters. The heart rate recovery coefficient in the mapping relationship is a key adjustment parameter, calculated by analyzing the correspondence between the user's historical heart rate variability sequence and exercise intensity sequence. Specifically, the system extracts the user's heart rate recovery curve after completing a standard exercise test, calculates the time required for the heart rate to drop from its peak to resting level, and compares it with a population benchmark to derive an individualized coefficient. For example, when the user's heart rate recovery time is significantly longer than the benchmark, the heart rate recovery coefficient is set to a lower value (e.g., 0.7). In this case, the physiological response mapping unit proportionally reduces the exercise intensity parameters in the initial plans, such as adjusting "30 minutes of moderate-intensity exercise" in the original plan to "40 minutes of low-intensity exercise."

[0037] The behavioral tolerance correction unit adjusts the sleep schedule parameters based on the sleep sensitivity coefficient. The sleep sensitivity coefficient is obtained by analyzing the correlation between the user's environmental exposure feature vector and the sleep quality sequence. The system establishes a regression model between temperature and humidity change sequences and sleep stage transition frequencies, calculating the amplitude of sleep quality fluctuations caused by unit changes in environmental parameters as the sensitivity coefficient. For example, when the user's sleep quality has a high sensitivity coefficient to temperature changes (e.g., a 1°C increase in temperature leads to a 15% reduction in deep sleep), the behavioral tolerance correction unit adjusts the initial plan's "go to bed 1 hour earlier" to "go to bed 30 minutes earlier," and adds environmental temperature control suggestions (e.g., "maintain a sleep environment of 22±1°C"). The correction process uses an iterative adjustment algorithm, re-evaluating the matching degree between the plan parameters and physiological response parameters after each adjustment until the following convergence condition is met:

[0038] in: This indicates the corrected intervention parameter value. This represents the theoretical parameter values ​​based on the physiological response model. This is a preset allowable deviation threshold (usually set to 0.1). When the adjustment deviations of all intervention parameters meet this condition, the correction process terminates, and a set of correction schemes containing physiological adaptation parameters is output.

[0039] The health plan optimization module receives the revised set of intervention intensity parameters and adjusts the plan based on the user's historical plan compliance records. The feature space construction unit organizes multi-dimensional features into a structured dataset, where each plan corresponds to a data point in the feature space. Dimensions include: exercise intensity correction coefficient (continuous value, range 0.5-1.5), sleep adjustment magnitude (discrete value: mild / moderate / significant), dietary intervention complexity (categorical value: simple / moderate / complex), and historical execution completion rate (percentage value). For example, the feature vector of a plan can be represented as [0.8,"moderate","moderate",75%], representing that the plan has moderate-intensity exercise adjustment, significant changes in sleep patterns, moderately complex dietary recommendations, and a historical execution rate of 75%.

[0040] The mapping adjustment unit uses a nearest neighbor propagation clustering algorithm to identify optimization directions in the feature space. The algorithm automatically determines the number and location of cluster centers, grouping data points with similar features into the same cluster. Each cluster center represents a type of health management plan with similar characteristics, and the distribution of plan completion rates around the cluster center reflects the feasibility of that type of plan. The system analyzes the statistical distribution of completion rates in each cluster and identifies the characteristic patterns of high-completion-rate clusters (e.g., clusters with completion rates ≥ 80%) as optimization directions. For example, if high-completion-rate clusters generally have the characteristics of "exercise intensity correction coefficient 0.7-0.9" and "simple dietary intervention complexity," the system adjusts plans with exercise intensity coefficients higher than 0.9 in the initial plan set to 0.85 and simplifies complex dietary recommendations into phased implementation. The adjustment process follows the gradient direction in the feature space, gradually migrating the plan feature vectors towards high-completion-rate cluster centers, ultimately generating an optimized set of health management plans.

[0041] The intervention parameters are individually adapted through a physiological response model, and then the optimal combination of features is found in the feature space based on historical compliance data. The physiological adaptation parameters established in the parameter coupling stage provide key inputs for the optimization stage, while the feature space mapping discovers feasible solution patterns through a data-driven approach. This dual optimization mechanism ensures both the scientific rigor and the feasibility of the plan, forming a closed-loop health management plan generation system.

[0042] Example 4: See Figure 5This embodiment involves the collaborative operation of a health pathway generation module and a health dynamic update module. The system receives a set of optimized health management plans from the health plan optimization module. This set contains multiple personalized health intervention plans optimized through parameter coupling and feature space optimization. The time-series pattern analysis unit in the health pathway generation module first analyzes the user's daily behavioral pattern characteristics and extracts key time feature parameters. These features are derived from the user's long-term behavioral log data, including activity records for weekdays and rest days. The system identifies the user's fixed activity periods each day, such as breakfast time from 7:00-8:00 AM, lunch break from 12:00-1:00 PM, dinner time from 6:00-7:00 PM, and leisure time from 8:00-10:00 PM. Simultaneously, the system marks behavioral transition nodes, such as the transition point from sleep to activity in the morning (usually 6:30-7:00 AM) and the pattern switching point from weekdays to rest days (Friday evening to Sunday evening). These nodes represent critical moments when the user's lifestyle rhythm changes.

[0043] Based on these temporal characteristics, the pathway optimization unit begins to optimize the temporal arrangement and combination of intervention measures. This unit matches the various intervention measures in the optimized health management plan with the distribution of users' free time slots, considering the required duration of each measure, the optimal implementation time, and the logical relationship between measures. For example, morning blood pressure monitoring should be scheduled within one hour of waking up, aerobic exercise is recommended to be scheduled two hours after meals, and sleep environment adjustments should be completed before bedtime. The system uses a constraint satisfaction algorithm, with the objective function of maximizing the effectiveness of measure implementation and minimizing schedule conflicts, to generate multiple candidate health management implementation pathway plans.

[0044] Table 1: Examples of Health Management Implementation Pathways

[0045] The health dynamic update module receives the latest feedback data from users in real time. This data includes real-time physiological indicators uploaded by wearable devices, dietary records entered by users through a mobile application, exercise completion status, and real-time data reported by environmental sensors. The feedback feature extraction unit analyzes key feature changes from these data streams, such as the deviation of morning blood pressure measurements from baseline, the difference between the actual and recommended duration of a midday walk, and adherence to carbohydrate intake at dinner. The system establishes a mapping relationship between these feature changes and health status labels, identifying which interventions have produced the expected effects and which measures need adjustment.

[0046] The incremental learning unit dynamically updates at each intervention node in the health management execution pathway. When the system detects that the execution effect of a certain intervention node deviates from expectations, it triggers a local adjustment mechanism. For example, if morning blood pressure values ​​do not show the expected improvement for three consecutive days, the system may adjust the blood pressure measurement frequency or increase monitoring parameters; if compliance with midday walks remains low, the system may suggest adjusting the exercise time or switching to an indoor exercise program. These adjustments not only target individual interventions but also consider the synergistic effects between measures, such as adjusting nutritional recommendations accordingly when increasing exercise intensity.

[0047] The update process employs a gradual optimization strategy, adjusting only one or a few intervention parameters at a time, observing the feedback after each adjustment, and then proceeding with further optimization. The system maintains a version history, tracking the reasons, content, and effects of each adjustment to form a complete optimization trajectory. This incremental update mechanism ensures that the health management execution pathway can adapt to the user's dynamically changing health status and lifestyle, while maintaining the integrity and continuity of the plan.

[0048] Through meticulous time-series planning and dynamic feedback adjustments, static health intervention plans are transformed into executable pathways that adapt to individual lifestyles. The health pathway generation module provides a structured execution framework, while the health dynamic update module adds adaptability and flexibility to this framework. The collaborative work of these two modules ensures that health management plans are not only scientifically sound but also highly feasible and sustainable in practice.

[0049] Example 5: This example involves the terminal processing flow of the health decision output module. The system receives the updated set of user health status tags from the health dynamic update module and the health management execution pathway plan output by the health pathway generation module. The plan integration unit initiates the feature association alignment process to match the two types of input data in multiple dimensions. This unit first parses the latest status identifiers in the updated set of user health status tags, such as dynamic tags like "morning blood pressure fluctuation risk" and "inadequate postprandial blood glucose control"; at the same time, it parses the sequence of intervention measures in the health management execution pathway plan, including the type, target parameters, and execution sequence of each measure. Feature association alignment is carried out based on three dimensions: time feature matching, intensity feature calibration, and target feature mapping. Time feature matching aligns the time attributes of the health status tags with the time nodes of the pathway plan. For example, the "morning blood pressure fluctuation risk" tag is associated with the blood pressure measurement measure at 06:45 in the pathway plan, and the "inadequate midday blood glucose control" tag is associated with the walking measure at 12:30. Intensity feature calibration adjusts the intensity parameters of the intervention measures according to the severity of the status tags. For example, when the risk level of the inadequate blood glucose control tag increases, the recommended duration of the midday walk is increased accordingly. Target feature mapping establishes a correspondence between state improvement goals and the effects of measures, such as mapping the improvement goal of blood pressure fluctuation risk label to the adjustment of bedroom temperature control parameters in environmental intervention.

[0050] The instruction generation unit constructs a structured instruction tree based on the correlation alignment results. This unit first transforms the sequence of measures in the health management execution pathway into a basic instruction framework, with each measure corresponding to a branch node in the instruction tree. Then, based on the feature correlation alignment results, the basic instructions undergo parameterization and logical enhancement. Parameterization is reflected in the dynamic adjustment of specific execution details. For example, the blood pressure measurement instruction dynamically sets the measurement frequency (increasing to twice daily when the risk is high) and additional monitoring items (adding pulse wave velocity detection) based on the risk level of morning blood pressure fluctuations. Logical enhancement is reflected in the explicit expression of dependencies between measures, such as embedding conditional judgment logic in the dinner nutrition intervention instruction: if the exercise intervention is not completed that day, the allowed carbohydrate intake is automatically reduced by 20%. The final generated instruction tree contains a multi-layered nested structure: the root node is the daily health management theme (e.g., "Cardiovascular Risk Control Day"), the first-level branches are time period blocks (morning, noon, evening), the second-level branches are specific intervention categories (monitoring, nutrition, exercise), and the leaf nodes are atomic instruction items with parameters.

[0051] Atomic instruction items encapsulate execution elements using a standardized data structure. Each instruction item contains six core fields: Instruction Type field defines the nature of the operation (monitoring / nutrition / exercise / environment / behavior), Execution Content field describes the specific action requirements (e.g., "perform diastolic blood pressure measurement"), Time Window field specifies the execution period (e.g., 06:45±15 minutes), Duration field limits the operation duration (5 minutes), Parameter Threshold field gives the target range (systolic blood pressure <135 mmHg), and Associated Device field lists the required tools (electronic blood pressure monitor model). Instruction items are related through logical connectors, such as sequential relationships (blood pressure measurement → breakfast intake), conditional relationships (activate an alert if systolic blood pressure >140), and mutual exclusion relationships (alternate between aerobic exercise and strength training every other day). The final health management recommended instructions are output in JSON-LD format, containing a complete instruction tree structure and metadata description, supporting machine parsing and visualization rendering.

[0052] The system pushes final instructions to a multi-terminal execution platform via a message queue. Upon receiving the instructions, the mobile application generates an interactive task list, triggers a pop-up reminder at a specified time, and guides the user to perform the operation. The smart home control terminal parses environmental intervention instructions and automatically adjusts air conditioning temperature parameters. Medical devices pre-start measurement programs based on monitoring instructions, reducing user operation steps. All terminal execution data is transmitted back to the system through a unified interface, forming a closed loop from decision generation to execution feedback. The instruction version management system records the differences in each instruction update, generating change description documents when health status changes significantly, helping users understand the logic behind the solution adjustments.

[0053] Through multi-level feature association and structured instruction generation, complex health management logic is transformed into actionable daily tasks. The dynamic mapping relationship established by the solution integration unit ensures real-time adaptation of measures and states, while the tree structure constructed by the instruction generation unit enables the visual expression and automated execution of management plans. This terminal output mechanism transforms the calculation results of the preceding modules into user-perceptible and executable health management guidelines, completing the value transformation chain of personalized health management from data analysis to behavioral intervention.

[0054] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0055] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A health management solution recommendation system based on big data analysis, characterized in that, include: The health data acquisition module is used to synchronously collect users' physiological indicator time-series data, health behavior time-series data, and environmental exposure time-series data, and to align the physiological indicator time-series data, health behavior time-series data, and environmental exposure time-series data with timestamps. The health feature quantification module is used to extract features from the time-series data of physiological indicators, health behavior, and environmental exposure output by the health data acquisition module after timestamp alignment, and generate physiological indicator feature matrix, health behavior quantification matrix, and environmental exposure feature vector, respectively. The health status identification module is used to classify the physiological indicator feature matrix, health behavior quantification matrix and environmental exposure feature vector output by the health feature quantification module based on the preset health status classification rules, and generate a set of user health status tags. The health plan prediction module is used to predict and generate an initial set of health management plans based on the set of user health status tags output by the health status recognition module, combined with a pre-stored health intervention knowledge base, and through an association rule mining algorithm. The health parameter coupling module is used to perform physiological adaptation correction on the intervention intensity parameters in the initial health management plan set output by the health plan prediction module, based on the mapping relationship between human physiological response parameters and behavioral intervention parameters.

2. The health management solution recommendation system based on big data analysis according to claim 1, characterized in that, The health data acquisition module includes: A multi-source data synchronization unit is used to synchronously collect the time-series data of physiological indicators, health behavior, and environmental exposure. The time-series data of physiological indicators includes heart rate variability and blood glucose concentration. The time-series data of health behavior includes exercise intensity and sleep quality. The time-series data of environmental exposure includes air quality index and temperature and humidity change. The timestamp alignment unit is used to perform timestamp alignment on the physiological indicator time-series data, health behavior time-series data, and environmental exposure time-series data output by the multi-source data synchronization unit using an interpolation compensation method. The data quality cleaning unit is used to remove outlier data points from the time-series data of physiological indicators, health behaviors, and environmental exposures output by the timestamp alignment unit, and generate a standardized health time-series dataset.

3. The health management solution recommendation system based on big data analysis according to claim 1, characterized in that, The health characteristic quantification module includes: The physiological indicator analysis unit is used to extract time-domain variability features and frequency-domain energy distribution features from the time-series data of physiological indicators in the standardized health time-series dataset, and generate the physiological indicator feature matrix. The behavior pattern quantification unit is used to extract the distribution features of behavior duration and the gradient features of intensity change from the health behavior time series data in the standardized health time series dataset, and generate the health behavior quantification matrix. The environmental exposure parsing unit is used to extract exposure frequency features and cumulative exposure amount features from the environmental exposure time series data in the standardized health time series dataset, and generate the environmental exposure feature vector.

4. The health management solution recommendation system based on big data analysis according to claim 1, characterized in that, The health status recognition module includes: The status classification unit is used to classify the physiological indicator feature matrix, health behavior quantification matrix and environmental exposure feature vector according to a preset metabolic syndrome risk level threshold, and generate a preliminary health status label. The tag fusion unit is used to analyze the correlation between tags based on the preliminary health status tags output by the state classification unit, and fuse them to generate the user health status tag set.

5. The health management solution recommendation system based on big data analysis according to claim 1, characterized in that, The health plan prediction module includes: The knowledge base matching unit is used to retrieve matching basic intervention item entries from the health intervention measure knowledge base based on the user health status tag set. The association rule mining unit is used to discover the combination patterns of measures based on the basic intervention measure items output by the knowledge base matching unit, and generate the initial health management plan set by using a frequent itemset mining algorithm.

6. The health management solution recommendation system based on big data analysis according to claim 1, characterized in that, The health parameter coupling module includes: The physiological response mapping unit is used to adaptively modify the exercise intensity parameters in the initial health management program set based on the heart rate recovery coefficient in the mapping relationship between the human physiological response parameters and behavioral intervention parameters. The behavioral tolerance correction unit is used to perform tolerance correction on the sleep sensitivity coefficient in the mapping relationship between the human physiological response parameters and behavioral intervention parameters, based on the sleep sensitivity coefficient.

7. The health management solution recommendation system based on big data analysis according to claim 1, characterized in that, The system further includes: a health plan optimization module, used to adjust the initial health management plan set based on the corrected intervention intensity parameters output by the health parameter coupling module, combined with the user's historical plan compliance records, using a feature space mapping method, to generate an optimized health management plan set; the health plan optimization module includes: The feature space construction unit is used to construct a multi-dimensional feature space by combining the modified intervention intensity parameter with the execution completion rate index in the user's historical plan compliance record. The mapping adjustment unit is used to identify the optimization direction in the multidimensional feature space generated by the feature space construction unit through the nearest neighbor propagation clustering algorithm, and adjust the initial health management scheme set to generate the optimized health management scheme set.

8. The health management solution recommendation system based on big data analysis according to claim 1, characterized in that, The system further includes: a health pathway generation module, used to generate a health management execution pathway plan by optimizing the set of optimized health management plans output by the health plan optimization module, combining user daily behavior pattern characteristics, and performing time-series arrangement and combination optimization of intervention measures; the health pathway generation module includes: The time-series pattern analysis unit is used to extract the idle time distribution features and behavior transition nodes from the user's schedule behavior pattern features; The pathway optimization unit is used to generate the health management execution pathway scheme based on the idle time period distribution characteristics and behavior transition nodes output by the time series pattern analysis unit, combined with the intervention measures in the optimized health management scheme set, through a constraint satisfaction algorithm.

9. The health management solution recommendation system based on big data analysis according to claim 1, characterized in that, The system further includes: a health dynamic update module, used to receive the latest user feedback data in real time, and update the user's health status label set based on the intervention nodes in the health management execution path plan output by the health path generation module through an incremental feature learning method; the health dynamic update module includes: The feedback feature extraction unit is used to extract physiological indicator shift features and behavioral compliance change features from the latest user feedback data. An incremental learning unit is used to incrementally update the user's health status label set at the intervention node of the health management execution pathway scheme based on the physiological index shift features and behavioral compliance change features output by the feedback feature extraction unit.

10. The health management solution recommendation system based on big data analysis according to claim 1, characterized in that, The system further includes: a health decision output module, used to generate a final health management recommendation instruction based on the updated user health status tag set output by the health dynamic update module and the health management execution path scheme output by the health path generation module; the health decision output module includes: The scheme integration unit is used to perform feature association and alignment between the updated user health status tag set and the health management execution path scheme. The instruction generation unit is used to generate the final health management recommendation instruction, which includes a sequence of intervention measures and execution time nodes, based on the correlation alignment results output by the scheme integration unit.

Images