A heart health management system and method based on an intelligent internet of things system

Abstract

The application relates to a heart health management system and method based on an intelligent Internet of Things system, and relates to the technical field of intelligent medical treatment and health management. The system adopts a modular architecture comprising data acquisition, preprocessing and feature engineering, intelligent risk grading evaluation, three-level health management closed-loop execution, model iteration and effect tracking, and early warning message pushing. Multidimensional health data is collected, cleaned, time-aligned and feature-screened, a neural network model based on BiLSTM and an attention mechanism is used to output a risk level, three-level health management strategies are dynamically matched to realize automatic intervention, and the model and rules are continuously optimized through effect feedback. The application can effectively solve the problems of difficult fusion of multi-source heterogeneous data, low risk evaluation precision and disconnection between monitoring and intervention, and realizes whole-process closed-loop management of heart health.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare and health management technology, and in particular to a cardiac health management system and method based on a smart Internet of Things system, which is applicable to multi-dimensional health monitoring, intelligent risk classification assessment and automated closed-loop intervention management for cardiac patients in home settings. Background Technology

[0002] Cardiovascular chronic diseases have become a major threat to residents' health, with heart disease receiving particular attention due to its high incidence and mortality rates. Current heart disease management primarily relies on a combination of regular checkups at medical institutions and home self-monitoring by patients. Technically, wearable devices or home medical instruments are typically used to collect users' physiological data such as heart rate and blood pressure, which is then transmitted to a cloud platform for storage and display. Some advanced systems incorporate data analysis algorithms to perform trend analysis on the collected historical data; when monitored values ​​exceed preset thresholds, the system sends an alert to the user or their family. Furthermore, some health management platforms offer online consultation services, allowing users to upload examination reports to obtain doctor's advice, forming a preliminary technical path of data collection – anomaly alerts – human consultation.

[0003] However, the main problems existing in the prior art include: (1) There are difficulties in the spatiotemporal fusion and feature extraction of multi-source heterogeneous health data, resulting in insufficient accuracy of risk assessment models and difficulty in capturing hidden temporal features before the onset of disease; (2) There is a lack of automated closed-loop connection between risk detection results and health management system, making it impossible to dynamically map risk levels with standardized graded intervention strategies; (3) There is a lack of automated execution solutions for the entire process from risk warning to personalized intervention and then to critical care, making it difficult to achieve deep collaboration between monitoring and management; (4) The existing system has insufficient privacy protection during the data collection process and relies heavily on user-initiated operations, making it difficult to achieve seamless continuous monitoring; (5) The lack of a standardized intervention strategy library for different risk levels leads to a lack of effective implementation plans after the early warning, especially the imperfect emergency response mechanism in high-risk situations. Summary of the Invention

[0004] The purpose of this invention is to provide a cardiac health management system and method based on an intelligent Internet of Things system, which overcomes the shortcomings of existing technologies such as difficulty in spatiotemporal fusion of multi-source heterogeneous health data, low accuracy of risk assessment, and disconnection between risk detection and the automated closed loop of the health management system, and realizes in-depth mining of multi-dimensional health data, accurate modeling, and closed-loop management of the entire process of prevention-conditioning-treatment.

[0005] The implementation process of this invention is as follows: A cardiac health management system based on an intelligent Internet of Things (IoT) system includes a data acquisition module, a data preprocessing and feature engineering module, a cardiac health risk intelligent grading and assessment module, a Mynm three-level health management closed-loop execution module, a model iteration and effect tracking module, and an early warning and message push module. Data acquisition module: used to acquire users' multi-dimensional health data; Data preprocessing and feature engineering module: used to preprocess and feature filter the health data acquired by the data acquisition module, and output a core multidimensional spatiotemporal feature set related to cardiac health risks; The intelligent cardiac health risk grading and assessment module is used to read the core feature set output by the data preprocessing and feature engineering module, and output the user's cardiac health risk level and quantitative risk report. The Mainm three-level health management closed-loop execution module is the core execution unit of the system. It dynamically maps the output results of the heart health risk intelligent classification and assessment module. It has a built-in standardized execution method library for three-level health management and implements intervention strategies for at-risk populations. Model Iteration and Effect Tracking Module: This module is used to collect changes in users' health data after intervention by the Maiem three-level health management closed-loop execution module, evaluate the intervention effect, and feed the effect data back to the heart health risk intelligent classification assessment module and the Maiem three-level health management closed-loop execution module to achieve continuous iterative optimization of the accuracy of the heart health management system and the intervention plan. Early warning and message push module: Used for pushing the user's heart health risk level and quantitative risk report output by the intelligent heart health risk grading assessment module, the intervention strategy output by the Mainm three-level health management closed-loop execution module, and emergency rescue information through multiple channels, including user terminal, family terminal, medical staff terminal and medical institution terminal.

[0006] Furthermore, the data acquisition module collects user health data in real time. After being processed by the data preprocessing and feature engineering module, the data is input into the heart health risk intelligent classification assessment module to complete the risk determination. Then, the intervention strategy is executed through the Mainm three-level health management closed-loop execution module. The system self-optimizes through the model iteration and effect tracking module, and information is pushed through the early warning and message push module to form a complete closed loop.

[0007] Furthermore, the multi-dimensional health data includes physiological indicator data, sleep structure data, and behavioral activity data. It also supports users to manually upload physical examination report data. During the model training phase, the collected multi-dimensional health data will be used as model input, and the data samples will be labeled with corresponding risk tags. During the prediction phase, the user's real-time multi-dimensional health data will be collected and input into the fully trained model to output risk prediction results.

[0008] Furthermore, the intelligent risk grading and assessment module for heart health incorporates a lightweight neural network model based on BiLSTM and attention mechanism. The activation function and parameters of the lightweight neural network model can be iteratively optimized. The lightweight neural network model uses bidirectional LSTM to mine the sequential correlation features of time series data and assigns higher weights to core risk features through the attention mechanism.

[0009] Furthermore, the standardized implementation method library for the three-level health management includes a first-level health management sub-module, a second-level health management sub-module, and a third-level health management sub-module, which correspond to the implementation of intervention strategies for low-, medium-, and high-risk groups, respectively.

[0010] The method for managing heart health using the above system includes the following steps: (1) Collect 24-hour continuous health data of users in the home scenario by the data acquisition module of the heart health management system based on the intelligent Internet of Things system; The health data includes data on sleep, heart rate, blood pressure, electrocardiogram, blood oxygen, and blood glucose collected by wearable terminals; sleep structure, sleep heart rate, and respiratory rate data collected by smart sleep belts; and activity level, circadian rhythm, and activity range data collected by home sensors. The data acquisition module synchronizes the data to edge computing nodes. All terminals have no image acquisition function, thus protecting user privacy throughout the process.

[0011] (2) The data preprocessing and feature engineering modules are used to clean, denoise and align the collected health data, and the data sampling frequency is uniformly set to 1 minute / time; the time features are encoded by sine and cosine coding, the continuous numerical data is standardized by min-max normalization, and the binary data is stored in 0 / 1 encoding form; the core features of heart health risk are screened by random forest feature selection algorithm, a multi-dimensional spatiotemporal feature set is constructed, and a dynamic health database for users is generated. The core characteristics of cardiac health risks include 12 key features such as heart rate variability, blood pressure diurnal rhythm, percentage of deep sleep, average daily activity level, and ST segment changes on electrocardiogram.

[0012] (3) Using the intelligent classification and assessment module for heart health risk, a lightweight heart health risk prediction model based on BiLSTM and attention mechanism is constructed. The multidimensional spatiotemporal feature set is input into the model. The model mines the sequential correlation features of time series data through the bidirectional BiLSTM layer, assigns higher weights to core risk features through the attention mechanism layer, and finally outputs the user's risk level through the fully connected layer and softmax activation function, which are divided into three categories: low risk, medium risk and high risk. At the same time, a quantitative health risk report is generated, marking abnormal indicators and risk factors. (4) Based on the risk level output by the model, the Maiem three-level health management closed-loop execution module is used to automatically match the Maiem three-level health management strategy; (5) Utilize the model iteration and effect tracking module to continuously collect health data after user intervention, evaluate the intervention effect monthly, and feed the effect data back to the heart health risk intelligent classification assessment module to complete the parameter iteration optimization of the system. At the same time, update the standardized execution method library of the three-level health management in the Maimu three-level health management closed-loop execution module to achieve continuous optimization of the intervention plan.

[0013] Furthermore, in step (2), the time features are encoded using sine and cosine coding to generate a 24-hour periodic time feature vector. The encoding formula is as follows: Where h represents hours, m represents minutes, and s represents seconds, the generated time feature vector ranges from [-1, 1], and then... Normalize to the interval [0,1].

[0014] Furthermore, in step (2), continuous numerical data is standardized using min-max normalization to normalize the data to the [0,1] interval. The normalization formula is: .

[0015] Furthermore, in step (3), the calculation formula for the BiLSTM layer is: in As input features, To update the door, For the Gate of Oblivion For output of the hidden layer, These are candidate memory states.

[0016] Furthermore, in step (4), if the risk is low, first-level health management is initiated, routine monitoring is continuously implemented, a health trend report is generated monthly, personalized diet, exercise, and rest guidance is pushed, and a comprehensive risk review is completed every quarter. If the risk level is medium, Level II health management will be initiated. A personalized treatment plan combining traditional Chinese and Western medicine will be generated for abnormal indicators. The changes in indicators will be tracked weekly, and community medical staff will conduct online follow-ups every two weeks. The treatment effect will be evaluated and the plan will be optimized monthly. If the risk level is high, a three-tiered health management system will be activated. Warning information will be immediately sent to users, emergency contacts, and community healthcare workers via APP, SMS, and telephone. At the same time, an emergency call terminal will be triggered, and emergency green channels of top-tier hospitals nationwide will be activated to complete expert consultation, registration, and referral within 10 minutes, minimizing the treatment cycle.

[0017] Furthermore, the personalized treatment plan combining traditional Chinese and Western medicine includes a daily exercise plan, dietary formula, work and rest adjustments, and nutritional intervention program.

[0018] The positive effects of this invention: (1) This application continuously collects multi-dimensional health raw data such as physiological indicators, sleep structure and behavioral activities through the data acquisition module, and constructs a comprehensive data foundation by combining manually uploaded physical examination report data; on this basis, the data preprocessing and feature engineering module is used to perform data cleaning, time alignment and standardized coding, and the core multi-dimensional spatiotemporal feature set is screened out through principal component analysis, correlation coefficient method and random forest feature selection algorithm, thereby effectively solving the feature extraction problem caused by different formats and large noise interference of multi-source heterogeneous data; (2) In this application, the intelligent classification and assessment module for heart health risk calls the built-in lightweight neural network model based on BiLSTM and attention mechanism, uses the bidirectional BiLSTM layer to mine the correlation features of time series data, and uses the attention mechanism layer to assign higher weights to the core risk features, thereby outputting accurate user heart health risk level and quantitative risk report, which significantly improves the accuracy of identifying hidden risks. (3) In this application, the Maim three-level health management closed-loop execution module dynamically maps to the first, second or third level health management sub-modules according to the assessment results, automatically executes the standardized intervention strategies for low, medium and high risk groups, and realizes the seamless connection from risk warning to personalized intervention; at the same time, the model iteration and effect tracking module collects the data changes after the intervention and feeds them back to the model and rule base, so that the system has the ability to continuously self-optimize. (4) The early warning and message push module in this application pushes key information to users, family members, medical staff and medical institutions through multiple channels to ensure rapid response in emergency situations; (5) This application effectively solves the core problems of existing technologies in multi-source data fusion, risk assessment accuracy and monitoring and intervention disconnection, realizes closed-loop management of the whole process of prevention-conditioning-treatment, and improves the systematicness, reliability and response efficiency of chronic disease management. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the heart health management system based on the intelligent Internet of Things system described in Example 1; Figure 2This is a hierarchy diagram of the lightweight neural network model in Example 1; Figure 3 This is a physical image of the system product and a schematic diagram of the equipment data type in Example 2; Figure 4 This is a schematic diagram of the home scene space layout in Example 2; Figure 5 This is a flowchart of the method for managing heart health using the system described in Example 1 in Example 3. Detailed Implementation

[0020] The technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] This application constructs a complete technical architecture integrating data collection, intelligent assessment, tiered intervention, and effect feedback. It not only achieves in-depth mining and precise modeling of multi-dimensional health data, but also establishes an automated dynamic mapping mechanism between risk levels and standardized intervention strategies, forming a logically rigorous and self-evolving health management closed loop. At the same time, it provides timely early warnings and message pushes, offering a systematic health management method for the early detection and scientific management of heart health.

[0022] This application constructs a spatiotemporal fusion and feature extraction method for multi-source heterogeneous health data, achieving synchronous collection, temporal alignment, and standardized fusion of multi-dimensional data on physiology, sleep, and behavior. This allows for in-depth mining of hidden temporal features prior to heart disease onset, solving the problems of insufficient accuracy and high false negative rates in traditional single-device indicator assessments. Furthermore, it pioneers a deeply coupled architecture between intelligent risk grading assessment for heart health and the Mainm three-tiered health management system, achieving automated dynamic mapping between risk levels and tiered intervention strategies. This forms a closed-loop management system encompassing prevention, management, and treatment, fundamentally reducing the risk of acute heart attacks and sudden cardiac death, addressing the core pain point of "disconnect between monitoring and management" in traditional models. Moreover, it integrates high-quality national elderly care resources and emergency green channels, enabling multi-terminal synchronous early warning and rapid dispatch of medical resources in high-risk situations. This eliminates the time spent queuing, registration, and referral, minimizing the response cycle for critical care and securing golden treatment time for patients, thus overcoming the deficiency of existing systems that lack a concrete treatment plan after early warning.

[0023] Example 1 A cardiac health management system based on an intelligent Internet of Things system, see Figure 1It includes a data acquisition module, a data preprocessing and feature engineering module, a cardiac health risk intelligent grading and assessment module, a Mainm three-level health management closed-loop execution module, a model iteration and effect tracking module, and an early warning and message push module. Data Acquisition Module: Used to acquire users' multi-dimensional health data; supports manual uploading of physical examination report data, and completes structured storage and access control of data; the multi-dimensional health data includes physiological indicator data, sleep structure data, and behavioral activity data. It also supports users manually uploading physical examination report data. During the model training phase, the collected multi-dimensional health data will be used as model input, and the data samples are labeled with corresponding risk tags. During the prediction phase, the user's real-time multi-dimensional health data is collected and input into the fully trained model, and the risk prediction results are output.

[0024] Data Preprocessing and Feature Engineering Module: Built-in data cleaning, time-series alignment, and standardized coding algorithms, as well as principal component analysis, mutual information, and random forest feature selection algorithms, are used to preprocess and feature filter the health data acquired by the data acquisition module, and output a core multidimensional spatiotemporal feature set related to heart health risks; The intelligent cardiac health risk grading and assessment module reads the core feature set output by the data preprocessing and feature engineering module, and outputs the user's cardiac health risk level and quantitative risk report. This module incorporates a lightweight neural network model based on BiLSTM and attention mechanisms. This lightweight neural network model includes an input layer, normalization layer, convolutional layer, feature extraction layer, Dropout regularization layer, BiLSTM and attention mechanism layer, connection layer, and output layer. See details... Figure 2 The input layer embeds vital sign trend data, activity trajectory data, lifestyle pattern data, and environmental safety data into the normalization layer. The normalization layer then passes the normalized data through a convolutional layer, a feature extraction layer, a Dropout regularization layer, a BiLSTM and attention mechanism layer, and a connection layer to output low-risk probability, medium-risk probability, high-risk probability, and probability of sudden events. The activation function and parameters of the lightweight neural network model can be iteratively optimized. The lightweight neural network model uses a bidirectional BiLSTM to mine the sequential correlation features of time-series data and assigns higher weights to core risk features through an attention mechanism.

[0025] The Mainm three-level health management closed-loop execution module is the core execution unit of the system. It dynamically maps the output results of the intelligent cardiac health risk grading assessment module and has a built-in standardized execution method library for three-level health management to implement intervention strategies for at-risk populations. The standardized execution method library for three-level health management includes a first-level health management sub-module, a second-level health management sub-module, and a third-level health management sub-module, which correspond to the implementation of intervention strategies for low, medium, and high-risk populations, respectively.

[0026] Model Iteration and Effect Tracking Module: This module is used to collect changes in users' health data after intervention by the Maiem three-level health management closed-loop execution module, evaluate the intervention effect, and feed the effect data back to the heart health risk intelligent classification assessment module and the Maiem three-level health management closed-loop execution module to achieve continuous iterative optimization of the accuracy of the heart health management system and the intervention plan. Early warning and message push module: Used for pushing the user's heart health risk level and quantitative risk report output by the intelligent heart health risk grading assessment module, the intervention strategy output by the Mainm three-level health management closed-loop execution module, and emergency rescue information through multiple channels, including user terminal, family terminal, medical staff terminal and medical institution terminal.

[0027] During system operation, the data acquisition module collects user health data in real time. After being processed by the data preprocessing and feature engineering module, the data is input into the intelligent cardiac health risk grading assessment module to complete the risk assessment. Then, the intervention strategy is executed through the Mainm three-level health management closed-loop execution module. The system self-optimizes through the model iteration and effect tracking module and pushes information through the early warning and message push module, forming a complete closed loop.

[0028] In this embodiment, the data acquisition module refers to a hardware and software combination unit responsible for collecting users' multi-dimensional raw health data. It works in conjunction with the data preprocessing and feature engineering module to transmit the collected raw data to subsequent processing stages. Specifically, the data types acquired by the data acquisition module can include physiological indicator data (such as heart rate, blood pressure, ECG, blood glucose, blood oxygen, etc.), sleep structure data (such as deep sleep percentage, respiratory rate, etc.), and behavioral activity data (such as activity level, circadian rhythm, etc.). Furthermore, this module can also support users manually uploading physical examination report data to supplement static health records. This module can consist of physical devices such as wearable terminals, smart sleep belts, home sensors, and smart gateways, and can also include a mobile application interface for receiving manually uploaded data. For example, this module can include devices such as a smart care host, a smart speaker, a water immersion sensor, a smart pillbox, an entry sensor, a door magnetic sensor, an air quality detector, an SOS emergency call device, a 6-in-1 vital signs detector, a smart sleep belt, a smart socket, a smart thermometer and hygrometer, and a toilet sensor. These devices converge data around the central gateway device. Alternatively, the module may include only a combination of key vital sign detection devices and environmental monitoring devices. This application does not impose any special limitations on this approach. The module continuously collects data in a home environment in a seamless manner, protecting user privacy throughout the process, without generating image data. The collected data stream directly drives subsequent preprocessing procedures.

[0029] In this embodiment, the data preprocessing and feature engineering module refers to the processing unit that cleans, aligns, encodes, and filters the original data. It works in conjunction with the data acquisition module to receive the original data and with the intelligent cardiac health risk grading and assessment module to output a core feature set. Specifically, this module internally deploys a data cleaning algorithm to remove outliers, a time-series alignment algorithm to unify the sampling frequency of different sensors (e.g., to 1 minute / sample), and a standardized encoding algorithm. Standardized encoding may include min-max normalization for continuous numerical data, sine and cosine encoding for time features, and 0 / 1 encoding for binary data. Furthermore, this module integrates principal component analysis, correlation coefficient methods, and random forest feature selection algorithms to filter out a core multidimensional spatiotemporal feature set strongly correlated with cardiac health risk from massive amounts of data, such as heart rate variability, blood pressure diurnal rhythm, deep sleep percentage, daily activity level, and ST segment changes in electrocardiogram. This module can be a software service cluster running on edge computing nodes or cloud servers. For example, this module can be a preprocessor deployed locally on the smart gateway, or it can be a distributed data processing platform in the cloud. This application does not impose any particular limitation on this. Through the processing of this module, heterogeneous data is transformed into standard feature vectors that the model can recognize, providing high-quality input for subsequent risk assessment.

[0030] In this embodiment, the intelligent cardiac health risk grading and assessment module refers to a computational unit that quantifies and assesses a user's cardiac health risk based on artificial intelligence algorithms. It works in conjunction with the data preprocessing and feature engineering module to read the core feature set and with the Mainm three-level health management closed-loop execution module, using its output as the trigger for intervention strategy execution. Specifically, this module incorporates a lightweight neural network model based on a BiLSTM + attention mechanism. The model's structure includes an input layer, convolutional layer, BiLSTM layer, attention mechanism layer, fully connected layer, and output layer. The input layer receives vital sign trend data, activity trajectory data, lifestyle pattern data, and environmental safety data. After processing through embedding and normalization layers, convolutional layers, feature extraction layers, Dropout regularization layers, LSTM / attention layers, and fully connected layers, the output layer finally outputs low-risk probability, medium-risk probability, high-risk probability, and probability of sudden events. The model's activation function and parameters can be iteratively optimized to adapt to individual differences among users. This module can be deployed on a high-performance computing cluster in the cloud or, after pruning and quantization, run as an inference engine at the edge. For example, this model can use bidirectional LSTM to mine the sequential correlation features of time-series data and assign higher weights to core risk features through an attention mechanism. This application does not impose any specific limitations on this aspect. This module captures hidden time-series features through deep learning technology, achieving intelligent mapping from data to risk levels.

[0031] In this embodiment, the Mainm three-level health management closed-loop execution module is the core execution unit that automatically executes standardized intervention strategies based on risk levels. It dynamically maps to the intelligent cardiac health risk grading and assessment module and works in conjunction with the early warning and message push module to implement interventions. This module has a built-in standardized execution rule library for three-level health management, divided into a first-level health management sub-module, a second-level health management sub-module, and a third-level health management sub-module, corresponding to intervention strategies for low, medium, and high-risk groups, respectively. Specifically, when a low-risk signal is received, the first-level health management sub-module initiates routine monitoring and health guidance; when a medium-risk signal is received, the second-level health management sub-module initiates personalized treatment plans and follow-ups; and when a high-risk signal is received, the third-level health management sub-module initiates emergency early warnings and medical resource allocation. This module can be a software system containing a rule engine and a workflow engine. For example, this module can automatically match corresponding service packages, such as dietary formulas, exercise plans, or emergency green channels, based on a preset rule library. It can also dynamically adjust the thresholds in the rule library based on user historical feedback. This embodiment does not impose any special limitations on this. This module ensures that users at different risk levels receive appropriate interventions, achieving a seamless integration of monitoring and management.

[0032] In this embodiment, the model iteration and effect tracking module refers to a feedback unit that collects post-intervention data and optimizes system performance. It works in conjunction with the Mainm three-level health management closed-loop execution module to collect changes in user health data after intervention, evaluate the intervention effect, and feed the effect data back to the heart health risk intelligent grading assessment module and the three-level management rule base. Specifically, this module continuously monitors the improvement of users' physiological indicators after implementing the intervention strategy, evaluating the effect by comparing the data differences before and after the intervention. If the effect is significant, the current strategy is strengthened; if the effect is poor, the model parameters are retrained or the rule base is updated. This module can be an independent data analysis service or embedded in the above modules as a background process. For example, this module can automatically generate an effect evaluation report monthly and send the labeled feature data back to the model training end. It can also monitor fluctuations in key indicators in real time, triggering a re-evaluation immediately upon detecting an abnormal rebound. This embodiment does not impose any special limitations on this. Through this module, the system possesses self-evolution capabilities, continuously improving prediction accuracy and service targeting as data accumulates.

[0033] In this embodiment, the early warning and message push module refers to a notification unit responsible for sending risk information and health guidance through multiple channels. It works in conjunction with the Mainm three-level health management closed-loop execution module to push content to designated terminals according to the instructions generated by the module. The pushed content includes risk warning information, health guidance plans, and emergency assistance information. Push terminals can include user terminals (such as mobile apps), family terminals, medical staff terminals, and medical institution terminals. Specifically, this module supports multiple communication protocols, such as SMS, telephone, and APP message push. In high-risk scenarios, this module can simultaneously trigger emergency call terminals and link with the emergency green channels of top-tier hospitals nationwide. This module can be a message middleware integrating multiple communication interfaces. For example, when a high risk is detected, this module can simultaneously send an alarm to the user's mobile phone, send a notification SMS to family members, send a work order to the community doctor, and automatically dial the emergency number. It can also push routine health tips to a single terminal. This embodiment does not impose any special limitations on this. This module ensures that key information can reach relevant responsible persons in a timely manner, shortening the emergency response time.

[0034] Example 2 Deploying a cluster of smart IoT devices in a user's home environment, see Figure 3-4 The system includes a smart sleep belt, a 6-in-1 vital signs detector, and a door magnetic sensor. The devices are connected to a cloud system via a smart gateway. The system automatically collects daily data on the user's vital signs (scientific sleep staging, sleep quality assessment, body temperature, heart rate, blood oxygen, blood pressure, blood sugar, uric acid, etc.) and activity trajectory data (nighttime - number of times getting out of bed; daytime - indoor footprints, outdoor - outdoor records, etc.), as well as daily routine data (eating, watching TV, cooking, laundry, washing, getting up at night, etc.) and environmental safety data (temperature, humidity, gas, TVOC, PM2.5, door opening and closing, floor water immersion, etc.).

[0035] The data preprocessing module aligns data from different frequencies to a 1-minute interval and uses a random forest algorithm to select 12 core features, including heart rate variability and ST segment changes.

[0036] The intelligent risk grading and assessment module for heart health loads a pre-trained BiLSTM model, inputs features, calculates that the user is currently in a medium-risk state, and generates a quantitative report that includes markers of abnormal blood pressure diurnal rhythm.

[0037] After receiving a medium-risk signal, the Maimu three-level health management closed-loop execution module automatically activates the second-level health management sub-module to generate a personalized plan that includes a low-salt diet formula, a daily 30-minute brisk walking plan, and suggestions for traditional Chinese medicine conditioning.

[0038] The alerts are sent to the user's app and the linked community doctor's app via the alert and push notification module. The community doctor conducts online follow-ups every two weeks through the system.

[0039] A month later, the model iteration and effect tracking module analysis found that the user's blood pressure index tended to stabilize. Therefore, the positive feedback data was marked and sent back to the training set to fine-tune the model parameters to enhance the recognition accuracy of this group of people. At the same time, the recommended threshold for exercise intensity in the rule base was updated.

[0040] This embodiment significantly improves the accuracy of cardiac health risk identification and early warning capabilities, while reducing false negative and false positive rates, due to the deep fusion and feature engineering optimization of multi-source heterogeneous data. The introduction of a lightweight neural network model based on BiLSTM+ attention mechanism effectively captures the dependencies between time-series data and assigns higher weights to key risk factors, enhancing the model's interpretability and predictive performance. The establishment of a three-tiered closed-loop management system—from low-risk prevention and medium-risk management to high-risk emergency care—automatically matches risk levels with intervention measures, breaking the bottleneck of traditional models that emphasize monitoring but neglect intervention. The configuration of a multi-terminal collaborative early warning and rapid medical resource dispatch mechanism greatly shortens the response time for critical care and improves the success rate of rescue. Furthermore, the system's continuous learning and self-optimization capabilities allow it to continuously improve personalized service capabilities as user data accumulates, making it suitable for various scenarios such as home-based elderly care and community-based health and wellness.

[0041] Example 3 The method for managing heart health using the system described in Example 1 includes the following steps: (1) Collect 24-hour continuous health data of users in the home scenario by the data acquisition module of the heart health management system based on the intelligent Internet of Things system; The health data includes data on sleep, heart rate, blood pressure, electrocardiogram, blood oxygen, and blood glucose collected by wearable terminals; sleep structure, sleep heart rate, and respiratory rate data collected by smart sleep belts; and activity level, circadian rhythm, and activity range data collected by home sensors. The data acquisition module synchronizes the data to edge computing nodes. All terminals have no image acquisition function, thus protecting user privacy throughout the process.

[0042] In this embodiment, the data acquisition terminal device is a fully contactless intelligent IoT data acquisition terminal, which refers to a collection of sensing devices deployed in the user's home environment that can operate automatically without the user's active intervention. This terminal specifically includes wearable devices, smart sleep bands, and distributed home sensors. Wearable devices typically exist in the form of wristbands or patches, used for continuous contact with the user's skin to collect physiological data such as heart rate, blood pressure, electrocardiogram waveform, blood oxygen saturation, and blood glucose levels. Smart sleep bands are laid under the mattress or integrated into bedding, used for non-contact capture of the user's body movement, respiratory rate, and sleep structure stage data during sleep. Home sensors are distributed in key areas such as the living room, bedroom, and bathroom to monitor the user's activity level, circadian rhythm, and activity range within the room. This data is transmitted to a smart gateway via wireless communication protocols such as ZigBee, Wi-Fi, or Bluetooth. The smart gateway, acting as an edge computing node, is responsible for receiving multi-source data streams and performing preliminary format verification and caching. All data acquisition terminals do not contain camera components, eliminating the generation of image data at the hardware source, thereby ensuring user privacy and security at the physical level. For example, when a user falls asleep at night, the smart sleep tracker records the number of times they turn over and the fluctuations in their breathing rate in real time. At the same time, wearable devices record nighttime heart rate variability, and home infrared sensors record the time and path of getting up at night. This data is aggregated at the local gateway to form an initial data packet. Through this seamless collection of multi-source heterogeneous data, a continuous, 24 / 7 user health profile can be built, avoiding data loss and subjective bias caused by traditional manual entry or intermittent measurement, and providing a high-fidelity data foundation for subsequent risk assessment.

[0043] (2) The data preprocessing and feature engineering modules are used to clean, denoise and align the collected health data, and the data sampling frequency is uniformly set to 1 minute / time; the time features are encoded by sine and cosine coding, the continuous numerical data is standardized by min-max normalization, and the binary data is stored in 0 / 1 encoding form; the core features of heart health risk are screened by random forest feature selection algorithm, a multi-dimensional spatiotemporal feature set is constructed, and a dynamic health database for users is generated. The core characteristics of cardiac health risks include 12 key features such as heart rate variability, blood pressure diurnal rhythm, percentage of deep sleep, average daily activity level, and ST segment changes on electrocardiogram.

[0044] Specifically, time features are encoded using sine and cosine coding to generate a 24-hour periodic time feature vector. The coding formula is as follows: Where h represents hours, m represents minutes, and s represents seconds, the generated time feature vector ranges from [-1, 1], and then... Normalize to the interval [0,1].

[0045] in, and It is a two-dimensional encoding result of the 24-hour cycle time feature, used to map "time of day" onto a two-dimensional plane to solve the problem of time cyclicality (for example, 23:59 and 00:01 are actually adjacent in time, but they would be very different if directly represented by numbers). It is the cosine encoded component of time in a two-dimensional plane (x-axis coordinate), given by the formula: The calculated value ranges from [-1, 1], which is used to represent the horizontal position of time on the "circular time axis" during the day.

[0046] It is the sinusoidal encoded component of time in a two-dimensional plane (y-axis coordinate), given by the formula: The calculated value ranges from [-1, 1], which is used to characterize the vertical position of time on the "circular time axis" during the day.

[0047] and Taken together, this maps a 24-hour day onto a unit circle, achieving a periodic representation of time. This allows the model to understand that 23:59 and 00:01 are close time points.

[0048] Continuous numerical data is standardized using min-max normalization, normalizing the data to the [0,1] interval. The normalization formula is as follows: .

[0049] In this embodiment, data cleaning and denoising refers to the process of removing outliers and noise data caused by sensor drift, signal interference, or transmission packet loss. This is typically achieved using moving average filtering or wavelet transform methods. Time alignment maps data from devices with different sampling frequencies (e.g., heart rate might be on the order of seconds, activity level on the order of minutes) onto a unified time axis. In this embodiment, all data is resampled at a frequency of 1 minute / time to ensure strict correspondence of each feature in the time dimension. Sine and cosine coding is a special processing method for time features. Because time has a periodicity (a 24-hour cycle), direct numerical conversion would break the periodic constraint. Therefore, hours, minutes, and seconds are converted into vectors with two dimensions: sine and cosine. The formula is... and This preserves the cyclical continuity of time. Minimum-maximum normalization is used to eliminate the influence of dimensional differences in different physiological indicators, mapping continuous values ​​such as blood pressure and blood glucose to the [0,1] interval. The calculation formula is as follows: Binary data, such as sensor on / off states, is directly encoded as 0 or 1. The random forest feature selection algorithm, based on the Gini impurity or information gain principle, evaluates the contribution of each feature to predicting cardiac health risks from a massive amount of raw features, automatically selecting the 12 most representative core features. These include heart rate variability reflecting autonomic nervous function, blood pressure diurnal rhythm reflecting cardiovascular load, the percentage of deep sleep reflecting recovery quality, daily average activity reflecting metabolic level, and ST segment changes in electrocardiogram reflecting myocardial ischemia. For example, when processing a user's data for one day, the system identifies a brief ST segment depression between 2 AM and 4 AM, accompanied by a deep sleep percentage below 15%. The random forest algorithm determines that this combination of features has a very high risk weight and includes it in the core feature set. Through the above standardized preprocessing and feature engineering, not only are the computational obstacles caused by data heterogeneity eliminated, but the dimensionality of the model input is also significantly reduced, improving the accuracy and convergence speed of subsequent risk assessments.

[0050] (3) Using the intelligent classification and assessment module for heart health risk, a lightweight heart health risk prediction model based on BiLSTM and attention mechanism is constructed. The multidimensional spatiotemporal feature set is input into the model. The model mines the sequential correlation features of time series data through the bidirectional BiLSTM layer, assigns higher weights to core risk features through the attention mechanism layer, and finally outputs the user's risk level through the fully connected layer and softmax activation function, which are divided into three categories: low risk, medium risk and high risk. At the same time, a quantitative health risk report is generated, marking abnormal indicators and risk factors. The calculation formula for the BiLSTM layer is as follows: in As input features, To update the door, For the Gate of Oblivion For output of the hidden layer, These are candidate memory states.

[0051] The intelligent risk grading and assessment module for heart health has a built-in lightweight neural network model based on BiLSTM and attention mechanism. The activation function and parameters of the lightweight neural network model can be iteratively optimized. The lightweight neural network model uses bidirectional BiLSTM to mine the correlation features between time series data and assigns higher weights to core risk features through the attention mechanism.

[0052] In this embodiment, BiLSTM (Bidirectional Long Short-Term Memory) is a deep learning architecture specifically designed for processing sequential data. It comprises two independent LSTM layers, forward and backward, capable of simultaneously utilizing past and future contextual information to understand the current state. This makes it particularly suitable for uncovering long-term physiological trends prior to heart disease onset. An attention mechanism layer is embedded after the BiLSTM layers to dynamically calculate the importance weights of each time step or feature in the input sequence. This allows the model to focus on the segments most critical to risk assessment, such as sudden spikes in blood pressure or persistent ST segment changes, while ignoring normal background fluctuations. Fully connected layers map the extracted high-dimensional temporal features to the sample label space, and the Softmax activation function transforms the output into a probability distribution belonging to low, medium, or high risk categories. The quantitative health risk report not only includes the final risk level conclusion but also details the specific abnormal indicators that led to the rating and their contribution, such as indicating that a 15% increase in the average nighttime heart rate over the past 7 days compared to baseline is a major risk factor. For example, when the model is input with a user's multidimensional spatiotemporal feature set for three consecutive days, the attention mechanism automatically assigns a high weight of 0.85 to the ST segment changes in the ECG during the early morning hours. Combined with blood pressure data from before and after these times, the model ultimately calculates a high-risk probability of 92% and generates a report suggesting suspected occult myocardial ischemia, recommending immediate follow-up. Through the synergistic effect of BiLSTM and the attention mechanism, the model captures both long-term pathological evolution patterns and accurately locates instantaneous critical signs, achieving intelligent and interpretable risk grading.

[0053] (4) Based on the risk level output by the model, the Maiem three-level health management closed-loop execution module is used to automatically match the Maiem three-level health management strategy; Specifically, it is based on users' long-term own data, takes the combination of traditional Chinese and Western medicine as the foundation, and medical services as the bottom-line guarantee to build a three-level closed loop of "data-conditioning-medical care". According to the output risk level, the assessment results are dynamically mapped with the self-developed Maiem three-level health management system, automatically triggering the standardized health management strategies of the corresponding level, realizing the whole-cycle, integrated scientific health management closed-loop intervention from health monitoring, precise conditioning to medical services.

[0054] If the risk level is low, Level 1 health management will be initiated. By setting up an IoT smart device at home, middle-aged and elderly people can continuously monitor their health, generate monthly health trend reports, push personalized guidance on diet, exercise, and rest, conduct a comprehensive risk review regularly, continuously track changes in key indicators, and generate health trend reports weekly, monthly, quarterly, and yearly. This enables the early detection of hidden risks and intervention in their nascent stages, thus blocking the progression of diseases from the source.

[0055] If the risk level is medium, initiate Level 2 health management. Based on the user's abnormal indicators and risk factors, and relying on the built-in integrated traditional Chinese and Western medicine intervention knowledge base, a personalized conditioning plan integrating traditional Chinese and Western medicine is generated. The personalized conditioning plan integrating traditional Chinese and Western medicine includes a daily exercise plan, diet formula, work and rest adjustment, and nutritional intervention plan. The conditioning effect and indicator changes are tracked weekly, and the plan is dynamically adjusted to reverse or delay the progression of risks and allow the body to return to a healthy balance. Community medical staff conduct online follow-ups every two weeks and evaluate the conditioning effect and optimize the plan monthly. If the risk level is high, initiate Level 3 health management. As the safety net of the entire system, the tertiary management medical service immediately triggers real-time risk warnings through multiple channels for health problems that do not improve after treatment. Warning information is immediately pushed to users, emergency contacts, and community healthcare workers via APP, SMS, and telephone. Simultaneously, the emergency call terminal is triggered, and the built-in national tertiary hospital medical resource database and emergency green channel are linked to complete expert connection, registration, referral scheduling and emergency resource linkage within 10 minutes, minimizing the response cycle for critical care and saving patients golden treatment time.

[0056] In this embodiment, the Mainm three-tier health management strategy is a set of tiered intervention execution rule bases dynamically bound to risk levels. Level 1 health management focuses on prevention and lifestyle guidance. Based on the user's daily activity data and preferences, the system automatically generates customized health recommendations, such as recommending low-salt diets or gentle exercises suitable for the elderly, and provides long-term trend feedback through monthly reports, aiming to maintain health and detect subtle changes early. Level 2 health management introduces professional medical intervention elements. For identified abnormal indicators, the system calls upon its built-in integrated traditional Chinese and Western medicine knowledge base to generate comprehensive conditioning plans covering drug assistance, dietary therapy, and exercise rehabilitation. A dual supervision mechanism of automatic system tracking and regular manual follow-up is established to ensure that intervention measures are implemented effectively and adjusted in a timely manner. Level 3 health management is an emergency response mechanism for critical situations. Once a high-risk condition is determined, the system immediately activates multi-channel alarm links, notifying not only the user but also simultaneously contacting family members, community doctors, and contracted hospitals. Using a preset emergency green channel interface, the system directly completes the registration, specialist appointment, and ambulance dispatch process in the background, eliminating cumbersome intermediate steps. For example, when the system determines a user is at high risk, it will send a warning text message to their children's mobile phones within one second, pop up an emergency pop-up window at the community doctor's workstation, and automatically call 120 (emergency services) to report the user's location and medical history. Simultaneously, it will lock in appointments with cardiology specialists at the hospital, ensuring the patient receives treatment within the golden timeframe. This tiered matching mechanism achieves the rational allocation of medical resources and the precise implementation of intervention measures, effectively solving the problem of the disconnect between early warning and treatment in the traditional model.

[0057] (5) Utilize the model iteration and effect tracking module to continuously collect health data after user intervention, evaluate the intervention effect monthly, and feed the effect data back to the heart health risk intelligent classification assessment module to complete the parameter iteration optimization of the system. At the same time, update the standardized execution method library of the three-level health management in the Maimu three-level health management closed-loop execution module to achieve continuous optimization of the intervention plan.

[0058] In this embodiment, intervention effect evaluation refers to comparing changes in key physiological indicators before and after the implementation of health management strategies to determine whether the risk level has decreased or stabilized. The system continuously collects new data from users after implementing diet, exercise, or drug treatments, performs differential analysis with the baseline data before intervention, and generates an effect evaluation report. If the data shows improvement in indicators, the current management strategy is strengthened; if the indicators do not improve or worsen, the rule base update mechanism is triggered to adjust the intensity or type of the intervention plan. Simultaneously, this new data labeled with intervention results is re-input into the training set of the risk prediction model. The weight parameters of the BiLSTM network are fine-tuned through the backpropagation algorithm, enabling the model to adapt to individual differences and changes in the disease spectrum, continuously improving prediction accuracy. For example, if a medium-risk user's blood lipid levels and blood pressure diurnal rhythm significantly return to normal after three months of a low-fat diet and aerobic exercise plan, the system marks this case as effective and increases the recommendation weight of such plans in similar populations; conversely, if the effect is unsatisfactory, the system will automatically try to replace it with other conditioning plans and record feedback. Through this closed-loop feedback and self-learning mechanism, the system has the ability to continuously evolve over time, ensuring the long-term effectiveness and adaptability of health management.

[0059] This application, through the synergistic effect of the above steps, constructs a complete technical closed loop from data acquisition, feature engineering, intelligent assessment to tiered intervention and effect iteration. The collaboration between the seamless data acquisition terminal and edge computing nodes resolves the contradiction between real-time acquisition of multi-source heterogeneous data and privacy protection, providing high-quality input for subsequent processing. The combination of sine and cosine encoding, normalization processing, and random forest feature selection effectively overcomes the spatiotemporal inconsistency of data, extracting high-value cardiac health risk features. The deep integration of BiLSTM and attention mechanisms significantly improves the ability to mine hidden temporal features and the accuracy of risk assessment. The dynamic mapping between the Maim three-level health management strategy and risk levels achieves a leap from passive monitoring to proactive intervention, especially the linkage of emergency green channels under high-risk conditions, which greatly shortens the treatment response time. The continuous effect data feedback mechanism ensures the self-evolution of the model and rule base, enabling the entire system to become increasingly accurate and intelligent with prolonged use, fundamentally reducing the risk of acute cardiac attacks and sudden cardiac death.

[0060] The core working principle of this application is as follows: through a seamless intelligent IoT terminal, multi-dimensional health data of users' physiology, sleep, and behavior are continuously collected in the home environment. Data preprocessing and feature fusion are completed through an edge computing and cloud collaboration architecture. Based on a large time series model, intelligent classification and assessment of cardiac health risks are realized. The risk level is then automatically matched with the Mainm three-level health management system to achieve a closed-loop management process of "early prevention for low risk, scientific management for medium risk, and rapid treatment for high risk".

[0061] The specific usage method is as follows: (1) Deploy a non-intrusive data collection terminal in the user's home environment. The system automatically completes 24-hour continuous health data collection without the need for active operation of the device. (2) The system automatically preprocesses, extracts features and assesses risks from the collected data. Users and their families can view health data, risk reports and health guidance plans in real time through the mini-program / APP. (3) When a user is in a low-risk state, the system automatically performs Level 1 health management and regularly pushes health living guidance and trend reports; When a user is in a medium-risk state, the system automatically implements level-two health management, pushes personalized scientific conditioning plans, and community medical staff follow up and conduct regular assessments of the conditioning effects. When a user is in a high-risk state, the system immediately triggers a multi-terminal emergency alert, simultaneously linking emergency contacts, medical personnel, and emergency green channels to quickly connect with high-quality medical resources and ensure the user's safety. (4) The system continuously tracks the intervention effect, automatically optimizes the risk assessment model and health management plan, and realizes long-term dynamic health management.

[0062] This application continuously collects multi-dimensional raw health data, including physiological indicators, sleep structure, and behavioral activities, through a data acquisition module, and combines this with manually uploaded physical examination reports to construct a comprehensive data foundation. Based on this, a data preprocessing and feature engineering module performs data cleaning, temporal alignment, and standardized coding. Principal component analysis, correlation coefficient method, and random forest feature selection algorithm are used to select a core multi-dimensional spatiotemporal feature set, effectively solving the feature extraction challenges caused by inconsistent formats and high noise interference from multi-source heterogeneous data. Subsequently, the intelligent cardiac health risk grading assessment module calls the built-in lightweight neural network model based on BiLSTM+ attention mechanism, utilizing bidirectional BiLSTM... The system mines the sequential correlation features of time-series data and assigns higher weights to core risk features using an attention mechanism layer, thereby outputting accurate user cardiac health risk levels and quantitative risk reports, significantly improving the accuracy of identifying hidden risks. Furthermore, the MyEM three-level health management closed-loop execution module dynamically maps assessment results to first-, second-, or third-level health management sub-modules, automatically executing standardized intervention strategies for low, medium, and high-risk groups, achieving seamless integration from risk warning to personalized intervention. Simultaneously, the model iteration and effect tracking module collects data changes after intervention and feeds them back to the model and rule base, enabling the system to continuously self-optimize. Finally, the warning and message push module pushes key information to users, families, medical staff, and medical institutions through multiple channels, ensuring rapid response in emergency situations. This effectively solves the core problems of existing technologies in multi-source data fusion, risk assessment accuracy, and the disconnect between monitoring and intervention, achieving a closed-loop management process encompassing prevention, conditioning, and treatment, and improving the systematic nature, reliability, and response efficiency of chronic disease management.

[0063] This application enables continuous, unobtrusive home-based monitoring of heart health and accurate early identification of hidden risks, addressing the core pain points of traditional models that rely on physical examinations and suffer from delayed early warnings. By quantifying users' health status through multi-dimensional data, it breaks the misconception that "no discomfort means health," achieving a fundamental shift from passive treatment to proactive prevention. Furthermore, it constructs the Mainm three-tiered health management closed-loop system, deeply integrating risk detection with scientific intervention and emergency treatment. This achieves a full-chain health protection system of "data-driven prevention, scientifically sound management, and expedited medical access," comprehensively reducing the risk of acute heart attacks and sudden cardiac death, minimizing or eliminating such tragedies. The entire system is replicable and scalable, widely applicable to various scenarios such as home-based elderly care, community-based elderly care, rehabilitation institutions, and corporate employee health management. It effectively improves the efficiency of chronic disease management and the quality of elderly care services, optimizes the allocation of medical resources, and reduces the burden on families and society caused by sudden serious illnesses.

[0064] Other embodiments of this application will readily conceive of by those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and embodiments are to be considered exemplary only, and this application is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A cardiac health management system based on an intelligent Internet of Things system, characterized in that: It includes modules for data acquisition, data preprocessing and feature engineering, intelligent risk assessment for heart health, closed-loop execution of the Mynm three-level health management system, model iteration and effect tracking, and early warning and message push. Data acquisition module: used to acquire users' multi-dimensional health data; Data preprocessing and feature engineering module: used to preprocess and feature filter the health data acquired by the data acquisition module, and output a core multidimensional spatiotemporal feature set related to cardiac health risks; The intelligent cardiac health risk grading and assessment module is used to read the core feature set output by the data preprocessing and feature engineering module, and output the user's cardiac health risk level and quantitative risk report. The Mainm three-level health management closed-loop execution module is the core execution unit of the system. It dynamically maps the output results of the heart health risk intelligent classification and assessment module. It has a built-in standardized execution method library for three-level health management and implements intervention strategies for at-risk populations. Model Iteration and Effect Tracking Module: This module is used to collect changes in users' health data after intervention by the Maiem three-level health management closed-loop execution module, evaluate the intervention effect, and feed the effect data back to the heart health risk intelligent classification assessment module and the Maiem three-level health management closed-loop execution module to achieve continuous iterative optimization of the accuracy of the heart health management system and the intervention plan. Early warning and message push module: used for the multi-channel push of user's heart health risk level and quantitative risk report output by the heart health risk intelligent classification assessment module, intervention strategies output by the Mainm three-level health management closed-loop execution module, and emergency rescue information.

2. The cardiac health management system based on an intelligent Internet of Things system according to claim 1, characterized in that: The data acquisition module collects user health data in real time. After being processed by the data preprocessing and feature engineering module, the data is input into the heart health risk intelligent classification assessment module to complete the risk assessment. Then, the intervention strategy is executed through the Mainm three-level health management closed-loop execution module. The system self-optimizes through the model iteration and effect tracking module, and information is pushed through the early warning and message push module to form a complete closed loop.

3. The cardiac health management system based on an intelligent Internet of Things system according to claim 1, characterized in that: The multi-dimensional health data includes physiological indicator data, sleep structure data, and behavioral activity data. It also supports users to manually upload physical examination report data. During the model training phase, the collected multi-dimensional health data will be used as model input, and the data samples will be labeled with corresponding risk tags. During the prediction phase, the user's real-time multi-dimensional health data will be collected and input into the fully trained model to output risk prediction results.

4. The cardiac health management system based on an intelligent Internet of Things system according to claim 1, characterized in that: The intelligent risk grading and assessment module for heart health has a built-in lightweight neural network model based on BiLSTM and attention mechanism. The activation function and parameters of the lightweight neural network model can be iteratively optimized. The lightweight neural network model uses bidirectional LSTM to mine the correlation features of time series data and assigns higher weights to core risk features through the attention mechanism.

5. The cardiac health management system based on an intelligent Internet of Things system according to claim 1, characterized in that: The standardized implementation method library for the three-level health management system includes a first-level health management sub-module, a second-level health management sub-module, and a third-level health management sub-module, which correspond to the implementation of intervention strategies for low-, medium-, and high-risk groups, respectively.

6. A method for managing heart health using the system of claim 1, characterized in that, Includes the following steps: (1) Collect 24-hour continuous health data of users in the home scenario by the data acquisition module of the heart health management system based on the intelligent Internet of Things system; (2) The data preprocessing and feature engineering modules are used to clean, denoise and align the collected health data, and the data sampling frequency is uniformly set to 1 minute / time; the time features are encoded by sine and cosine coding, the continuous numerical data is standardized by min-max normalization, and the binary data is stored in 0 / 1 encoding form; the core features of heart health risk are screened by random forest feature selection algorithm, a multi-dimensional spatiotemporal feature set is constructed, and a dynamic health database for users is generated. (3) Using the intelligent classification and assessment module for heart health risk, a lightweight heart health risk prediction model based on BiLSTM and attention mechanism is constructed. The multidimensional spatiotemporal feature set is input into the model. The model mines the sequential correlation features of time series data through the bidirectional BiLSTM layer, assigns higher weights to core risk features through the attention mechanism layer, and finally outputs the user's risk level through the fully connected layer and softmax activation function, which are divided into three categories: low risk, medium risk and high risk. At the same time, a quantitative health risk report is generated, marking abnormal indicators and risk factors. (4) Based on the risk level output by the model, the Maiem three-level health management closed-loop execution module is used to automatically match the Maiem three-level health management strategy; (5) Utilize the model iteration and effect tracking module to continuously collect health data after user intervention, evaluate the intervention effect monthly, and feed the effect data back to the heart health risk intelligent classification assessment module to complete the parameter iteration optimization of the system. At the same time, update the standardized execution method library of the three-level health management in the Maimu three-level health management closed-loop execution module to achieve continuous optimization of the intervention plan.

7. The method according to claim 6, characterized in that, In step (2), the time features are encoded using sine and cosine coding to generate a 24-hour periodic time feature vector. The encoding formula is as follows: Where h represents hours, m represents minutes, and s represents seconds, the generated time feature vector ranges from [-1, 1], and then... Normalize to the interval [0,1].

8. The method according to claim 6, characterized in that: In step (2), continuous numerical data is standardized using min-max normalization to normalize the data to the [0,1] interval. The normalization formula is: 。 9. The method according to claim 6, characterized in that: In step (3), the calculation formula for the BiLSTM layer is: in As input features, To update the door, For the Gate of Oblivion For output of the hidden layer, These are candidate memory states.

10. The method according to claim 6, characterized in that: In step (4), if the risk is low, initiate Level 1 health management, continue routine monitoring, generate a health trend report every month, push personalized diet, exercise, and rest guidance, and complete a comprehensive risk review every quarter. If the risk level is medium, Level II health management will be initiated. A personalized treatment plan combining traditional Chinese and Western medicine will be generated for abnormal indicators. The changes in indicators will be tracked weekly, and community medical staff will conduct online follow-ups every two weeks. The treatment effect will be evaluated and the plan will be optimized monthly. If the risk level is high, a three-tiered health management system will be activated. Warning information will be immediately sent to users, emergency contacts, and community healthcare workers via APP, SMS, and telephone. At the same time, an emergency call terminal will be triggered, and emergency green channels of top-tier hospitals nationwide will be activated to complete expert consultation, registration, and referral within 10 minutes, minimizing the treatment cycle.

Images