An intelligent health interpretation system based on multi-device IoT data fusion

Abstract

This invention discloses an intelligent health interpretation system based on multi-device IoT data fusion, belonging to the field of intelligent health management technology. The system includes a data acquisition and standardization module, a data fusion engine, a structured health knowledge base, an AI deep analysis module, a health status assessment module, and an intelligent interpretation and feedback module. It can connect to multiple heterogeneous IoT health devices, standardize the raw data, and then fuse it into a multi-dimensional dataset using user ID and timestamp as the core. Combining the detailed medical indicator thresholds and association rules in the knowledge base, AI algorithms are used to mine deep health correlations in the data, identify abnormal risks, construct a multi-dimensional assessment system, quantify health levels, and ultimately generate personalized health reports. This invention achieves a closed-loop health management system of data integration, deep analysis, accurate assessment, and personalized guidance, helping to transform from passive data viewing to proactive health prevention and improving the accuracy and effectiveness of health management.

Description

Technical Field

[0001] This invention relates to the field of intelligent health management technology, and in particular to an intelligent health interpretation system based on multi-device IoT data fusion. Background Technology

[0002] With the rapid penetration of IoT technology into the field of health management, various smart health monitoring devices have been widely adopted in homes and personal lives. These include smart bracelets for recording heart rate, steps, and sleep data; smart scales for measuring weight and body fat percentage; smart blood pressure monitors for monitoring systolic and diastolic blood pressure; and smart blood glucose meters for detecting fasting and postprandial blood glucose. These devices, with their convenient usage, can help users collect basic personal health data in real time, providing preliminary data support for daily health management and promoting the initial transformation of health management from passive medical treatment to proactive prevention.

[0003] However, existing technical solutions generally have the following shortcomings:

[0004] The problem of data silos is serious. Data formats are inconsistent and interfaces are not open between different brands and types of IoT devices, making it difficult for users to integrate data from different devices into a unified platform for centralized management and analysis.

[0005] The data interpretation capabilities are limited; most devices can only provide single-dimensional data displays and simple trend analysis, lacking the ability to comprehensively interpret multi-dimensional and cross-device data. For example, heart rate data alone cannot fully assess a user's fatigue level; it needs to be combined with data such as sleep quality and activity level.

[0006] AI analysis capabilities are limited. Existing AI analysis is mostly based on data from a single device or simple rules, making it difficult to capture the deep health correlations and potential risks that can only be revealed after data from multiple devices are integrated.

[0007] The lack of sufficient dimensions in health assessment makes it impossible to achieve a multi-dimensional and refined assessment of users' health status, which greatly reduces the pertinence and effectiveness of health recommendations.

[0008] Therefore, this application proposes an intelligent health interpretation system based on multi-device IoT data fusion. Summary of the Invention

[0009] One objective of this invention is to propose an intelligent health interpretation system based on multi-device IoT data fusion. This invention can break down data silos among heterogeneous IoT health devices. Through dedicated adapters and standardized processing, it converts and integrates data from various devices such as wristbands and blood pressure monitors into a unified format. Relying on a health knowledge base and AI, it combines detailed medical thresholds with algorithms to mine deep data correlations, identify risks, construct a multi-dimensional health assessment system and quantify levels, generate personalized reports, and provide feedback through multiple channels. The entire process ensures data security, provides closed-loop intelligent health management services, and helps transform from passively viewing data to proactive health prevention.

[0010] According to an embodiment of the present invention, an intelligent health interpretation system based on multi-device IoT data fusion includes a data acquisition and standardization module, a data fusion engine, a structured health knowledge base, an AI deep analysis module, a health status assessment module, and an intelligent interpretation and feedback module.

[0011] The data acquisition and standardization module is used to start the data acquisition service according to user authorization or preset plan, scan and connect at least four kinds of heterogeneous IoT health monitoring devices, collect raw health data after identifying the device type and model, and then convert the raw data of different formats and different communication protocols into a preset standardized data format through the built-in device-specific adapter and data converter.

[0012] The data fusion engine receives the standardized data and uses the user identifier as the core association key and the timestamp as the time sequence association basis to associate and merge data from different devices that reflect different health dimensions of the same user in the same time period or in continuous time periods, forming a multi-dimensional associated dataset.

[0013] The structured health knowledge base stores massive amounts of structured health information, including medical indicator thresholds categorized by age and gender, disease and symptom association rules, relationship between behavioral patterns and health impacts, health assessment rules and diagnostic logic developed by medical experts, and cross-indicator associations.

[0014] The AI ​​deep analysis module uses machine learning and deep learning algorithms to query the structured health knowledge base in real time during the analysis process, and calls on the medical knowledge and rules in it to guide the model learning and reasoning. It performs deep analysis on the fused multi-dimensional related dataset, specifically including identifying data time-series trends, detecting outliers, mining potential health risks and multi-indicator correlations.

[0015] The health status assessment module, based on the results of AI deep analysis and combined with the preset assessment model in the structured health knowledge base, quantifies and scores the user's health status in multiple dimensions such as cardiovascular health, metabolic health, sleep quality, and activity level, and calculates the comprehensive health level according to the weight of each dimension.

[0016] The intelligent interpretation and feedback module transforms the health status assessment results into personalized content that users can understand, including health reports, potential risk warning information, and actionable health improvement suggestions, and provides feedback to users through preset channels.

[0017] Furthermore, the data acquisition and standardization module supports at least four types of heterogeneous IoT health monitoring devices, including smart bracelets, smart blood pressure monitors, smart scales, and smart blood glucose meters, and can also be expanded to support smart sleep mats and smart exercise equipment.

[0018] The module supports one or more of the following communication protocols: Bluetooth, Wi-Fi, NFC, and Zigbee. It can also expand its protocol adaptation capabilities to new devices through firmware updates.

[0019] Furthermore, the data acquisition and standardization module adopts a unified standardization format, namely the HL7 FHIR industry standard format, a custom JSON format, or an XML format.

[0020] The standardization process specifically includes converting all timestamps to ISO8601 format, unifying measurement units according to preset rules, structuring data into key-value pairs of “user ID-device type-indicator name-indicator value-timestamp-data status”, and marking and excluding invalid data.

[0021] Furthermore, the association and merging process of the data fusion engine is specifically as follows:

[0022] First, filter all standardized data of the same user by UserID, then sort them by timestamp in time sequence, and determine the data of different devices with a time deviation within ±30 minutes as "data of the same time period" and perform dimensional correlation.

[0023] For data over a continuous time period, data is aggregated in one-hour time windows, and the average and extreme values ​​of equipment indicators within the window are merged to form a time-series multi-dimensional fusion dataset.

[0024] Furthermore, the AI ​​algorithms employed by the AI ​​deep analysis module include:

[0025] The Prophet algorithm for LSTM and ARIMA used in time series trend analysis;

[0026] Isolation Forest and One-Class SVM algorithms for anomaly detection;

[0027] Random Forest and Gradient Boosting SVM algorithms for classification and risk assessment;

[0028] Graph neural networks and association rule mining for multi-indicator association mining.

[0029] Furthermore, the medical indicator thresholds in the structured health knowledge base are subdivided by age group and gender: the normal resting heart rate range for women aged 18-44 is 60-70 beats / minute, the normal fasting blood glucose range for men aged 45-59 is 3.9-6.1 mmol / L, and the normal systolic blood pressure range for women aged 60 and above is 90-149 mmHg.

[0030] The cross-indicator correlation information specifically includes: when the percentage of deep sleep is <25%, the probability of an elevated resting heart rate increases by 30%;

[0031] If you walk less than 5,000 steps a day, the risk of having a body fat percentage that exceeds the normal range increases by 25%.

[0032] If a person's blood glucose level is >7.8 mmol / L 2 hours after a meal and their sleep duration is <6 hours, the risk of elevated systolic blood pressure the next day increases by 40%.

[0033] Furthermore, the health status assessment module employs a weighted model, a Bayesian network model, or a personalized model based on user historical data, and the assessment process specifically includes:

[0034] For the cardiovascular health dimension, blood pressure (weight 30%), heart rate (weight 25%), exercise duration (weight 20%), and sleep quality (weight 25%) are used as core indicators. The scores of each indicator are calculated according to the knowledge base threshold and then weighted and summed to obtain a dimension score of 0-100.

[0035] For the metabolic health dimension, the core indicators are blood glucose (weighted at 35%), weight (weighted at 25%), body fat percentage (weighted at 25%), and diet-related data (weighted at 15%). Similarly, the dimension score is calculated.

[0036] The sleep quality dimension uses sleep duration (weighted at 30%), deep sleep percentage (weighted at 40%), and number of turns (weighted at 30%) as indicators, while the activity level dimension uses daily steps (weighted at 40%), exercise intensity (weighted at 30%), and exercise frequency (weighted at 30%) as indicators.

[0037] The user's overall health score is calculated from 0 to 100, with a weighting of 30% for cardiovascular health, 30% for metabolic health, 20% for sleep quality, and 20% for activity level. The score is then divided into levels: 85-100 is excellent, 70-84 is good, 55-69 is average, and <55 is poor.

[0038] Furthermore, the preset feedback channels of the intelligent interpretation and feedback module include mobile APP, web portal, smart speaker, SMS push, and smart wearable device;

[0039] The intelligent interpretation and feedback module uses natural language generation technology to generate health reports, which specifically include indicator interpretation, risk analysis, and improvement suggestions.

[0040] Furthermore, it also includes a privacy and security protection module;

[0041] The privacy and security protection module anonymizes device identifiers and user identity information during the data acquisition phase. During the data transmission phase, it uses the TLS 1.3 protocol for end-to-end encryption to prevent theft or tampering during data transmission. During the data storage phase, it uses the AES-256 encryption algorithm to encrypt and store the fused data and evaluation results. During the data usage phase, the AI ​​analysis only calls the anonymized dataset and does not associate it with the user's identifiable identity information.

[0042] Furthermore, the structured health knowledge base has a regular update mechanism and a medical expert verification mechanism. The regular update mechanism is set to be once per quarter, and the updated content includes the latest medical guidelines and authoritative clinical research results. The medical expert verification mechanism is jointly completed by two general practitioners or relevant specialists with more than 5 years of clinical experience. The verification content includes the rationality of indicator thresholds and the accuracy of disease association rules. The verification results are recorded in writing and archived.

[0043] The AI ​​deep analysis module integrates LIME and SHAP model interpretability technologies. It generates a local interpretation report through LIME, showing the impact weight of individual indicators such as sleep duration and steps on cardiovascular health scores; and quantifies the contribution of each indicator to the overall health level through SHAP values.

[0044] The intelligent interpretation and feedback module can generate daily dietary suggestions, exercise plans, and daily routines based on the user's overall health level, shortcomings in various dimensions, and personal preferences. It also sends weekly reminders of the plan's progress via a mobile app and generates monthly adjustments to the plan based on the latest health data.

[0045] The beneficial effects of this invention are:

[0046] 1. The data acquisition and standardization module of this invention has a built-in device-specific adapter that can scan and connect to at least four types of mandatory heterogeneous IoT health monitoring devices and extended devices. It supports access via multiple communication protocols such as Bluetooth, Wi-Fi, NFC, and Zigbee. It can also expand its compatibility with new devices through firmware updates, breaking through the limitations of device brands and models. Secondly, the data acquisition and standardization module uses a data converter to uniformly convert raw data of different formats into the HL7 FHIR industry standard format, custom JSON format, or XML format. It simultaneously completes the unification of timestamps and measurement units, as well as the structured processing of "user ID-device type-indicator name-indicator value-timestamp-data status", eliminating invalid data. The data fusion engine then uses the user ID as the core association key and the timestamp as the time sequence basis to associate and merge the data of different devices into a multi-dimensional associated dataset. This not only realizes the dimensional association of data from multiple devices in the same time period, but also aggregates the mean and extreme values ​​of data in continuous time periods by 1-hour windows. Finally, it integrates the scattered device data into a unified data resource that can be centrally managed and deeply analyzed, completely solving the pain point of users having difficulty integrating data across devices.

[0047] 2. In this invention, multi-dimensional data covering cardiovascular, metabolic, sleep, and activity are first obtained through data fusion. Then, cross-indicator association rules and disease-symptom association rules stored in the structured health knowledge base are combined to make the interpretation no longer limited to a single indicator. Finally, the intelligent interpretation and feedback module uses natural language generation technology to transform the analysis results into personalized content that includes indicator interpretation, risk analysis, and improvement suggestions. For example, by combining heart rate data with sleep quality and activity data, it can be interpreted that a slightly high heart rate may be related to insufficient deep sleep and low daily steps, rather than simply stating that the heart rate is abnormal. This multi-dimensional and cross-device comprehensive interpretation can more comprehensively reveal the health meaning behind the data and solve the problem that a single-dimensional interpretation cannot reflect the whole picture of health.

[0048] 3. The AI ​​deep analysis module in this invention analyzes multi-dimensional correlated datasets after the fusion of multiple devices, rather than single-device data. This lays the foundation for capturing cross-device correlations. Furthermore, the AI ​​deep analysis module integrates multiple professional algorithms, using time-series trend analysis to track long-term changes in indicators, anomaly detection to identify outliers, and graph neural networks and association rules to mine hidden correlations among multiple indicators. More importantly, during the analysis process, it queries a structured health knowledge base in real time, calling medical indicator thresholds and expert rules to guide reasoning, avoiding misjudgments driven by pure data. Compared to AI analysis based on single data or simple rules, the AI ​​module of this application can accurately capture deep health correlations that only emerge after the fusion of multi-device data, identify potential health risks in advance, and significantly improve the depth and accuracy of the analysis.

[0049] 4. This invention breaks down health status into four core dimensions: cardiovascular health, metabolic health, sleep quality, and activity level. Each dimension selects and weights core indicators to avoid the one-sidedness of single-dimensional assessment. Medical indicator thresholds are subdivided by age and gender, and the assessment standards are tailored to the physiological characteristics of different populations. It also supports switching between weighted models, Bayesian network models, and personalized models. The personalized model can adjust indicator weights based on users' historical data to achieve personalized solutions for each individual. Finally, a comprehensive health score and level are calculated based on the weights of the four dimensions at 30%, 30%, 20%, and 20%, respectively. This multi-dimensional and refined assessment can objectively reflect the user's overall health status and provide a basis for generating targeted health recommendations, solving the problem of insufficient assessment dimensions leading to generalized and low-effectiveness recommendations. Attached Figure Description

[0050] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0051] Figure 1 This is a system architecture diagram of an intelligent health interpretation system based on multi-device IoT data fusion proposed in this invention;

[0052] Figure 2 This is a flowchart illustrating the implementation steps of an intelligent health interpretation system based on multi-device IoT data fusion proposed in this invention. Detailed Implementation

[0053] To make the technical means and objectives and effects of the present invention easier to understand, the embodiments of the present invention will be described in detail below with reference to specific illustrations.

[0054] like Figure 1-2 As shown, this invention discloses an intelligent health interpretation system based on multi-device IoT data fusion, including a data acquisition and standardization module, a data fusion engine, a structured health knowledge base, an AI deep analysis module, a health status assessment module, and an intelligent interpretation and feedback module;

[0055] The data acquisition and standardization module is used to start the data acquisition service according to user authorization or preset plan, scan and connect at least four kinds of heterogeneous IoT health monitoring devices, collect raw health data after identifying the device type and model, and then convert the raw data of different formats and different communication protocols into a preset standardized data format through the built-in device-specific adapter and data converter.

[0056] Specifically, heterogeneous IoT health monitoring devices include smart bracelets, smart blood pressure monitors, smart scales, and smart blood glucose meters, and can also be extended to connect to smart sleep mats and smart exercise bikes;

[0057] The data acquisition and standardization module supports communication protocols including Bluetooth 5.3, Wi-Fi 6, NFC, and Zigbee 3.0. To adapt to new devices, the device's proprietary communication protocol description file can be imported through firmware updates to expand compatibility.

[0058] Secondly, in the data standardization process, the unified format can be selected from the HL7 FHIR industry standard format, a custom JSON format, or an XML format. Specific processing steps include:

[0059] All timestamps of data collected from all devices are uniformly converted to ISO8601 format, measurement units are standardized according to preset rules, and the data is structured into key-value pairs of "User ID-Device Type-Indicator Name-Indicator Value-Timestamp-Data Status". Invalid data is marked and excluded, and does not proceed to subsequent processes. Taking a custom JSON format as an example, the standardized data is shown below:

[0060] {

[0061] "userID":"U002",

[0062] "deviceType":"Smart Blood Glucose Meter",

[0063] "indexName":"Postprandial 2-hour blood glucose",

[0064] "indexValue":6.8,

[0065] "unit":"mmol / L",

[0066] "timeStamp":"2024-06-15T13:30:20Z",

[0067] "dataStatus":"Valid"

[0068] }

[0069] The data fusion engine receives standardized data and uses user identifiers as the core association key and timestamps as the time sequence association basis to associate and merge data from different devices that reflect different health dimensions of the same user in the same time period or consecutive time periods, forming a multi-dimensional associated dataset.

[0070] Specifically, the association and merging process of the data fusion engine needs to be based on user identifiers and timestamps as the time sequence basis. The specific operation steps are as follows:

[0071] First, filter out all standardized data for the user by UserID, and sort them in ascending order by the timestamp field, such as from "2024-06-15T06:30:00Z" to "2024-06-15T18:30:00Z";

[0072] Next, data from different devices with a time deviation within ±30 minutes are classified as "data from the same time period" and correlated dimensionally. For example, the heart rate data of the smart bracelet collected by U002 at "2024-06-15T07:30:00Z" and the data of the smart blood pressure monitor (130 / 85mmHg) collected at "2024-06-15T07:45:00Z" are correlated as "heart rate-blood pressure" two-dimensional data from the same time period due to a time deviation of 15 minutes.

[0073] For data spanning consecutive time periods, aggregation is performed in hourly windows, merging the average and extreme values ​​of device metrics within each window. Taking the window "2024-06-15T08:00:00Z-2024-06-15T09:00:00Z" as an example, the smart bracelet reported steps three times, with an average of (1200+1350+1400) / 3≈1317 steps. The extreme values ​​were a minimum of 1200 steps and a maximum of 1400 steps. Simultaneously, blood pressure and blood glucose data within this window are integrated to form a time-series multi-dimensional fusion dataset, and data integrity is marked. For example, "complete" indicates that data from all four required devices were collected, while "partially missing" indicates that blood glucose meter data is missing.

[0074] The structured health knowledge base stores massive amounts of structured health information, including medical indicator thresholds categorized by age and gender, disease and symptom association rules, the relationship between behavioral patterns and health impacts, health assessment rules and diagnostic logic developed by medical experts, and cross-indicator associations.

[0075] Specifically, the medical indicator thresholds in the massive structured health information are subdivided by age group and gender. Specifically, they include the normal resting heart rate range of 60-70 beats / minute for women aged 18-44, the normal fasting blood glucose range of 3.9-6.1 mmol / L for men aged 45-59, and the normal systolic blood pressure range of 90-149 mmHg for women aged 60 and above. It also supplements data such as the normal resting heart rate of 55-65 beats / minute for men aged 18-44 and the normal fasting blood glucose of 3.8-6.0 mmol / L for women aged 45-59, to ensure coverage of different population groups.

[0076] Among them, the cross-indicator correlation information includes "the probability of increased resting heart rate increases by 30% when the percentage of deep sleep is <25%", "the risk of abnormal body fat percentage increases by 25% when the number of daily steps is <5000", and "the risk of increased systolic blood pressure the next day increases by 40% when the 2-hour postprandial blood glucose is >7.8mmol / L and the sleep duration is <6 hours". It also includes correlation rules such as "the risk of abnormal fasting blood glucose increases by 40% when the body fat percentage is >30%". All information comes from authoritative clinical research.

[0077] Secondly, the knowledge base also has a regular update mechanism and a medical expert verification mechanism. The regular update mechanism is set to be updated once a quarter, and the latest content is obtained by connecting to the PubMed database and the National Health Commission's "Medical Guidelines" column.

[0078] The verification mechanism is jointly completed by two general practitioners with more than 5 years of clinical experience. The verification content includes the rationality of the indicator thresholds and the accuracy of the disease association rules. If the two have different opinions, a third cardiovascular specialist will be invited to arbitrate. The verification results are archived in the format of "Expert Name-Review Date-Verification Content-Review Opinion" and kept for no less than 3 years.

[0079] The AI ​​deep analysis module employs machine learning and deep learning algorithms to query a structured health knowledge base in real time during the analysis process. It calls upon the medical knowledge and rules within the base to guide the model's learning and reasoning, and performs in-depth analysis on the fused multi-dimensional related dataset. Specifically, this includes identifying data time-series trends, detecting outliers, uncovering potential health risks, and identifying the correlation between multiple indicators.

[0080] Specifically, the AI ​​deep analysis module employs various machine learning and deep learning algorithms, combined with a structured health knowledge base, to conduct analysis. For time-series trend analysis, it uses the LSTM and ARIMA Prophet algorithms. Taking U002's heart rate data from the past 30 days as input, the LSTM model's input dimension is set to 30×1, outputting the predicted heart rate trend for the next 7 days. The Prophet algorithm then incorporates the "weekday / weekend" periodic factor to correct the trend. For example, if U002 takes more steps and has a slightly higher heart rate on weekends, Prophet can identify this pattern. At the same time, it queries the knowledge base to determine whether the prediction results deviate from the normal range for U002, a 45-year-old male, who falls within the normal heart rate range for the 45-59 age group.

[0081] For anomaly detection, the Isolation Forest and One-Class SVM algorithms are used, with a 95% confidence interval set. If U002's fasting blood glucose is 7.2 mmol / L and falls outside the confidence interval, it is judged as an outlier. At the same time, the rule "fasting blood glucose > 7.0 mmol / L may indicate prediabetes" in the knowledge base is called to mark the potential risk.

[0082] For classification and risk assessment, an SVM algorithm combining Random Forest and Gradient Boosting is used. Using U002's blood pressure, heart rate, exercise duration, and sleep quality as features, the algorithm outputs low, medium, and high cardiovascular health risk levels. The algorithm uses the evaluation weights from the knowledge base to adjust the importance of features during the process.

[0083] For multi-indicator association mining, graph neural networks and association rule mining are used. Combined with the rule "the effect of postprandial blood glucose and sleep duration on systolic blood pressure" in the knowledge base, hidden associations in the U002 data are mined. For example, U002 repeatedly shows "postprandial blood glucose 8.0 mmol / L + sleep 5.5 hours", and the average systolic blood pressure increases by 12 mmHg the next day, which verifies that the association rule is valid.

[0084] In addition, the module integrates LIME and SHAP model interpretability technologies. It generates local interpretation reports through LIME, such as "In the U002 cardiovascular health score, sleep duration reduces the score by 5 points because deep sleep accounts for only 22%". It quantifies the contribution of each indicator to the overall health level through SHAP value, such as "The blood glucose indicator SHAP value -2.3 positively improves the overall level; the sleep indicator SHAP value +1.8 negatively lowers the level".

[0085] Based on this, the health status assessment module conducts quantitative assessments by combining AI analysis results with preset assessment models in the knowledge base;

[0086] Specifically, the evaluation model can be a weighted model, a Bayesian network model, or a personalized model based on user historical data. The default is a weighted model. The specific evaluation process is as follows:

[0087] The cardiovascular health dimension uses blood pressure (30%), heart rate (25%), exercise duration (20%), and sleep quality (25%) as core indicators. The scores of each indicator are calculated according to the knowledge base threshold. For example, U002's average blood pressure is 130 / 85 mmHg (normal, score 100 points), and the weighted score is 100 × 30% = 30 points.

[0088] Heart rate average 65 beats / minute (normal, 100 points), weighted 100 × 25% = 25 points; Daily exercise duration 50 minutes (30-59 minutes range, 80 points), weighted 80 × 20% = 16 points; Sleep quality (deep sleep percentage 22%, 40 points), weighted 40 × 25% = 10 points.

[0089] The total score for each dimension is 30+25+16+10=81 points.

[0090] The metabolic health dimension is weighted as follows: 35% for blood glucose (U002 fasting blood glucose 5.8mmol / L, normal score 100 points, weighted 35 points), 25% for weight (BMI 22.5, normal score 100 points, weighted 25 points), 25% for body fat percentage (24%, normal score 100 points, weighted 25 points), and 15% for dietary data (daily vegetable intake 350g, meeting the target score 100 points, weighted 15 points). The total score is 35+25+25+15=100 points.

[0091] The sleep quality dimension (sleep duration of 6.2 hours: 70 points × 30% = 21 points; deep sleep percentage of 22%: 40 points × 40% = 16 points; number of turns over 10 times: 100 points × 30% = 30 points; total score of 67 points) and the activity level dimension (daily steps of 9,000: 90 points × 40% = 36 points; moderate exercise intensity: 100 points × 30% = 30 points; exercise frequency of 5 times per week: 100 points × 30% = 30 points; total score of 96 points).

[0092] The overall health score is calculated based on a weighted average of 30% for cardiovascular health, 30% for metabolism, 20% for sleep, and 20% for activity: 81×30%+100×30%+67×20%+96×20%=24.3+30+13.4+19.2=86.9 points. According to the grading standard (85-100 points is excellent), the overall health level is determined to be "excellent".

[0093] If the user selects the personalized model, the system will recalculate the score based on U002's health data from the past year. For example, if the duration of exercise has a significantly greater impact on the cardiovascular score than that of ordinary users, the weight of exercise duration will be adjusted to 25%, the weight of sleep quality will be adjusted to 20%, and the score will be recalculated.

[0094] At the same time, the intelligent interpretation and feedback module transforms the evaluation results into personalized content that users can understand. Preset feedback channels include mobile apps, web portals, smart speakers, SMS push notifications, and smart wearable devices.

[0095] Specifically, the intelligent interpretation and feedback module uses natural language generation technology to generate health reports, which include indicator interpretation, risk analysis, and improvement suggestions.

[0096] In addition, the intelligent interpretation and feedback module will generate daily dietary suggestions, exercise plans and sleep plans based on the user's overall health level and shortcomings. It will also push a reminder of the plan's progress every Sunday at 8:00 PM via the mobile app. On the last day of each month, it will generate plan adjustment suggestions based on the latest health data. For example, if your deep sleep percentage increases to 24% this month, it is recommended to adjust your sleep goal to 6.8 hours next month and add one yoga session per week to improve sleep quality.

[0097] It is worth mentioning that the privacy and security protection module must be implemented throughout the entire data processing workflow:

[0098] During the data acquisition phase, device identifiers, such as the smart blood pressure monitor serial number "OM-U30-654321", are anonymized as "Device_BP_002". User identity information is encrypted using the SHA-256 hash algorithm, and only the encrypted UserID (U002) is retained for data association.

[0099] During the data transmission phase, end-to-end encryption is performed using the TLS 1.3 protocol, and insecure protocols such as SSLv3 and TLS 1.0 / 1.1 are disabled to prevent data from being stolen or tampered with.

[0100] During the data storage phase, the fused data and evaluation results are encrypted and stored using the AES-256 encryption algorithm. Key management adopts a master key and data key mode. The master key is generated by the system administrator in the offline hardware encryption machine and is rotated once every 90 days. The data key is bound to the user data and is stored after being encrypted with the master key.

[0101] During the data usage phase, the AI ​​analysis module only calls the anonymized dataset. When medical experts review abnormal data, they only provide anonymous information such as "UserID:U002, indicator: postprandial blood glucose 8.0mmol / L, time: 2024-06-15", without associating it with identity details, to ensure user privacy and security.

[0102] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart health interpretation system based on multi-device IoT data fusion, characterized in that, It includes a data collection and standardization module, a data fusion engine, a structured health knowledge base, an AI deep analysis module, a health status assessment module, and an intelligent interpretation and feedback module; The data acquisition and standardization module is used to start the data acquisition service according to user authorization or preset plan, scan and connect multiple heterogeneous IoT health monitoring devices, collect raw health data after identifying the device type and model, and then convert the raw data of different formats and different communication protocols into a preset standardized data format through the built-in device-specific adapter and data converter. The data fusion engine receives the standardized data and uses the user identifier as the core association key and the timestamp as the time sequence association basis to associate and merge data from different devices that reflect different health dimensions of the same user in the same time period or in continuous time periods, forming a multi-dimensional associated dataset. The structured health knowledge base stores massive amounts of structured health information, including medical indicator thresholds categorized by age and gender, disease and symptom association rules, relationship between behavioral patterns and health impacts, health assessment rules and diagnostic logic developed by medical experts, and cross-indicator associations. The AI ​​deep analysis module uses machine learning and deep learning algorithms to query the structured health knowledge base in real time during the analysis process, and calls on the medical knowledge and rules in it to guide the model learning and reasoning. It performs deep analysis on the fused multi-dimensional related dataset, specifically including identifying data time-series trends, detecting outliers, mining potential health risks and multi-indicator correlations. The health status assessment module, based on the results of AI deep analysis and combined with the preset assessment model in the structured health knowledge base, quantifies and scores the user's health status in multiple dimensions such as cardiovascular health, metabolic health, sleep quality, and activity level, and calculates the comprehensive health level according to the weight of each dimension. The intelligent interpretation and feedback module transforms the health status assessment results into personalized content that users can understand, including health reports, potential risk warning information, and actionable health improvement suggestions, and provides feedback to users through preset channels.

2. The intelligent health interpretation system based on multi-device IoT data fusion according to claim 1, characterized in that, The data acquisition and standardization module supports heterogeneous IoT health monitoring devices, including smart bracelets, smart blood pressure monitors, smart scales, and smart blood glucose meters, and can also be expanded to support smart sleep mats and smart exercise equipment. The data acquisition and standardization module supports one or more of the following communication protocols: Bluetooth, Wi-Fi, NFC, and Zigbee. It can also expand its protocol adaptation capabilities to new devices through firmware updates.

3. The intelligent health interpretation system based on multi-device IoT data fusion according to claim 1, characterized in that, The data acquisition and standardization module adopts the following unified standardization formats: HL7 FHIR industry standard format, custom JSON format, and XML format. The standardization process specifically includes converting all timestamps to ISO8601 format, unifying measurement units according to preset rules, structuring data into key-value pairs of "user ID-device type-indicator name-indicator value-timestamp-data status", and marking and excluding invalid data.

4. The intelligent health interpretation system based on multi-device IoT data fusion according to claim 1, characterized in that, The association and merging process of the data fusion engine is as follows: First, filter all standardized data of the same user by UserID, then sort them by timestamp in time sequence, and determine the data of different devices with a time deviation within ±30 minutes as "data of the same time period" for dimensional correlation; For data over a continuous time period, data is aggregated in one-hour time windows, and the average and extreme values ​​of equipment indicators within the window are merged to form a time-series multi-dimensional fusion dataset.

5. The intelligent health interpretation system based on multi-device IoT data fusion according to claim 1, characterized in that, The AI ​​algorithms used in the AI ​​deep analysis module include: The Prophet algorithm for LSTM and ARIMA used in time series trend analysis; Isolation Forest and One-Class SVM algorithms for anomaly detection; Random Forest and Gradient Boosting SVM algorithms for classification and risk assessment; Graph neural networks and association rule mining for multi-indicator association mining.

6. The intelligent health interpretation system based on multi-device IoT data fusion according to claim 1, characterized in that, The medical indicator thresholds in the structured health knowledge base are subdivided by age group and gender. The normal resting heart rate range for women aged 18-44 is 60-70 beats / minute, the normal fasting blood glucose range for men aged 45-59 is 3.9-6.1 mmol / L, and the normal systolic blood pressure range for women aged 60 and above is 90-149 mmHg. The cross-indicator correlation information specifically includes: when the percentage of deep sleep is <25%, the probability of an elevated resting heart rate increases by 30%; If you walk less than 5,000 steps a day, the risk of having a body fat percentage that exceeds the normal range increases by 25%. If a person's blood glucose level is >7.8 mmol / L 2 hours after a meal and their sleep duration is <6 hours, the risk of elevated systolic blood pressure the next day increases by 40%.

7. The intelligent health interpretation system based on multi-device IoT data fusion according to claim 1, characterized in that, The health status assessment module uses a weighted model, a Bayesian network model, or a personalized model based on user historical data as its assessment model, and the assessment process is as follows: For the cardiovascular health dimension, blood pressure (weight 30%), heart rate (weight 25%), exercise duration (weight 20%), and sleep quality (weight 25%) are used as core indicators. The scores of each indicator are calculated according to the knowledge base threshold and then weighted and summed to obtain a dimension score of 0-100. For the metabolic health dimension, the core indicators are blood glucose (weighted at 35%), weight (weighted at 25%), body fat percentage (weighted at 25%), and diet-related data (weighted at 15%). Similarly, the dimension score is calculated. The sleep quality dimension uses sleep duration (weighted at 30%), deep sleep percentage (weighted at 40%), and number of turns (weighted at 30%) as indicators, while the activity level dimension uses daily steps (weighted at 40%), exercise intensity (weighted at 30%), and exercise frequency (weighted at 30%) as indicators. The user's overall health score is calculated from 0 to 100, with a weighting of 30% for cardiovascular health, 30% for metabolic health, 20% for sleep quality, and 20% for activity level. The score is then divided into levels: 85-100 is excellent, 70-84 is good, 55-69 is average, and <55 is poor.

8. The intelligent health interpretation system based on multi-device IoT data fusion according to claim 1, characterized in that, The preset feedback channels for the intelligent interpretation and feedback module include mobile APP, web portal, smart speaker, SMS push and smart wearable device; The intelligent interpretation and feedback module uses natural language generation technology to generate health reports, which specifically include indicator interpretation, risk analysis, and improvement suggestions.

9. The intelligent health interpretation system based on multi-device IoT data fusion according to claim 1, characterized in that, It also includes a privacy and security protection module; The privacy and security protection module anonymizes device identifiers and user identity information during the data acquisition phase. During the data transmission phase, it uses the TLS 1.3 protocol for end-to-end encryption to prevent theft or tampering during data transmission. During the data storage phase, it uses the AES-256 encryption algorithm to encrypt and store the fused data and evaluation results. During the data usage phase, the AI ​​analysis only calls the anonymized dataset and does not associate it with the user's identifiable identity information.

10. The intelligent health interpretation system based on multi-device IoT data fusion according to claim 1, characterized in that, The structured health knowledge base has a regular update mechanism and a medical expert verification mechanism. The regular update mechanism is set to be once a quarter, and the update content includes the latest medical guidelines and authoritative clinical research results. The medical expert verification mechanism is jointly completed by two general practitioners or relevant specialists with more than 5 years of clinical experience. The verification content includes the rationality of indicator thresholds and the accuracy of disease association rules. The verification results are recorded in writing and archived. The AI ​​deep analysis module integrates LIME and SHAP model interpretability technologies. It generates a local interpretation report through LIME, showing the impact weight of individual indicators such as sleep duration and steps on cardiovascular health scores; and quantifies the contribution of each indicator to the overall health level through SHAP values. The intelligent interpretation and feedback module can generate daily dietary suggestions, exercise plans, and daily routines based on the user's overall health level, shortcomings in various dimensions, and personal preferences. It also sends weekly reminders of the plan's progress via a mobile app and generates monthly adjustments to the plan based on the latest health data.

Images