Intelligent wearable health monitoring method and system based on artificial intelligence

Abstract

The invention relates to the technical field of health monitoring, in particular to an intelligent wearable health monitoring method and system based on artificial intelligence, and the method comprises the steps: collecting multi-modal physiological parameters and behavior data of a user through an intelligent wearable device, constructing a nonlinear mapping relation between health indexes and behavior data, and forming a personalized health state model; and based on the model, optimizing and solving a health early-warning threshold adaptive to the individual user, generating a target early-warning parameter and configuring the target early-warning parameter to the equipment, thereby realizing real-time dynamic monitoring and abnormal early-warning of the physiological parameters. In daily monitoring, physiological and behavior data are analyzed through trends, and intervention suggestions are generated for abnormal fluctuation or potential risks; and meanwhile, user historical health data and a medical diagnosis result are integrated, a health state model is corrected by using big data analysis, and the modeling precision is improved. And in combination with different user group characteristics, a personalized health knowledge base and a case base are constructed, a standardized health assessment scheme is formed, and the accuracy and the intelligent level of personalized health management are improved.

Description

Technical Field

[0001] This invention relates to the field of health monitoring technology, and in particular to an intelligent wearable health monitoring method and system based on artificial intelligence. Background Technology

[0002] With the widespread adoption of smart wearable devices, wearable technology-based health monitoring has become a crucial direction for achieving personalized health management. Traditional health monitoring methods often rely on single physiological parameters or fixed thresholds for health assessment, making it difficult to comprehensively integrate the intrinsic correlation between users' multimodal physiological data and behavioral habits. This results in a lack of dynamism and personalization in health status analysis. Furthermore, existing early warning mechanisms typically employ standardized thresholds, ignoring individual user differences and the impact of historical health data, leading to insufficient accuracy in early warnings. Moreover, their ability to analyze trends in long-term monitoring data and intervene in potential health risks is limited, hindering the formation of a closed-loop health management system. In addition, the health characteristics of different user groups vary significantly, and traditional methods lack the ability to build knowledge bases and iterate models based on group data, failing to achieve a combination of standardization and personalization in health assessment programs.

[0003] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main objective of this invention is to provide an intelligent wearable health monitoring method and system based on artificial intelligence, which aims to solve the technical problem that traditional health monitoring methods rely on single physiological parameters or fixed thresholds for health assessment, making it difficult to fully integrate the intrinsic correlation between users' multimodal physiological data and behavioral habits, resulting in a lack of dynamism and personalization in health status analysis.

[0005] To achieve the above objectives, the present invention provides an intelligent wearable health monitoring method based on artificial intelligence, the method comprising: By acquiring users' multimodal physiological parameters and behavioral data through smart wearable devices, determining users' health indicators based on the multimodal physiological parameters, and establishing a nonlinear mapping relationship between the users' health indicators and the behavioral data, a health status model is obtained. Based on the constructed health status model, the personalized health warning thresholds for target users are optimized and solved to obtain target warning parameters applicable to the health status of target users; The optimized target warning parameters are configured into smart wearable devices to collect and dynamically monitor the physiological parameters of target users in real time, and trigger a health risk warning mechanism when the physiological parameters are abnormal. During routine monitoring, trend analysis is performed on the physiological parameters and behavioral data of target users, and health intervention suggestions are generated for abnormal fluctuations or potential health risks. Acquire users' historical health data and medical diagnosis results, conduct big data analysis on physiological parameters, exercise behavior data and historical health data, and continuously revise and improve the health status model based on the inherent relationship between the data. To acquire health status models under the characteristics of different user groups, establish a health knowledge base and case library for personalized health monitoring, and form a standardized health assessment program; The platform integrates user health monitoring data, early warning records, and intervention effects for comprehensive management. It tracks changes in user health status in real time, assesses key health indicators and potential risks, and generates health reports based on the monitoring results.

[0006] Optionally, the multimodal physiological parameters include heart rate, blood oxygen saturation, and skin temperature; the behavioral data includes steps, exercise type, exercise intensity, and sleep structure; the process of acquiring the user's multimodal physiological parameters and behavioral data through a smart wearable device, determining the user's health indicators based on the multimodal physiological parameters, and establishing a nonlinear mapping relationship between the user's health indicators and the behavioral data to obtain a health status model includes: The user's multimodal physiological parameters and behavioral data are acquired through smart wearable devices, and the multimodal physiological parameters and behavioral data are preprocessed, including data cleaning, normalization and feature extraction. Based on the preprocessed multimodal physiological parameters, user health indicators were determined, including cardiovascular health score, metabolic health score and sleep health score; A health status model is obtained by establishing a nonlinear mapping relationship between the user's health indicators and the behavioral data through a long short-term memory network algorithm.

[0007] Optionally, the health warning thresholds include a resting heart rate abnormality threshold, an upper limit for active heart rate, and a lower limit for blood oxygen saturation. The optimization of the personalized health warning thresholds for the target user based on the constructed health status model employs a genetic algorithm-based intelligent optimization algorithm, which includes the following steps: According to the preset chromosome encoding method, the three physiological parameters, namely the abnormal resting heart rate threshold, the upper limit of active heart rate, and the lower limit of blood oxygen saturation, are encoded into binary strings, and the length of the binary strings is set according to the parameter accuracy requirements. Obtain the predefined binary string, set the fitness function of the genetic algorithm, and use a normalization method to convert the behavioral data into a dimensionless index; After the fitness function is set, a roulette wheel selection operation is used to select excellent individuals, and the crossover probability of the roulette wheel selection operation is set to a preset value. Gene recombination is carried out using a single-point crossover method. Based on the selected superior individuals, the genetic algorithm is run iteratively, and the population size of the genetic algorithm is set to the first preset value, and the maximum number of iterations is set to the second preset value. Once the genetic algorithm reaches the convergence condition or the maximum number of iterations, the personalized health warning threshold for the target user is decoded and used as the target warning parameter applicable to the target user's health status. The fitness function formula for the personalized health warning threshold for the target user is as follows: ; In the formula, For the fitness function, This is the threshold for abnormal resting heart rate. This is the upper limit of the active heart rate. This is the lower limit of blood oxygen saturation. The weight of the k-th health indicator satisfies , Based on the current threshold The output value of the health status model. These are the actual measured health indicator values. This represents the theoretical maximum value of the k-th health indicator. This represents the theoretical minimum value of the k-th health indicator.

[0008] Optionally, after the genetic algorithm reaches the convergence condition or the maximum number of iterations, decodes and obtains the personalized health warning threshold for the target user, and uses it as a target warning parameter applicable to the target user's health status, the method further includes: The multimodal physiological parameters are encoded as parameter combinations for individuals, and an initial population is randomly generated. The fitness of each individual is calculated based on the health status model, and superior individuals are selected and new offspring individuals are generated through crossover mutation. Through multiple iterations of evolutionary operations and optimization of parameter combinations, the individual with the highest fitness is finally selected, and its corresponding parameter combination is used as the target warning parameter.

[0009] Optionally, the step of configuring the optimized target warning parameters into the smart wearable device, collecting and dynamically monitoring the target user's physiological parameters in real time, and triggering a health risk warning mechanism when the physiological parameters are abnormal includes: The optimized target warning parameters are obtained and sent to the local processor of the smart wearable device via Bluetooth Low Energy protocol. Multi-sensor fusion technology is used to continuously collect users' heart rate and blood oxygen saturation data, and particle filtering algorithm is used to obtain the optimal estimated physiological parameters of the target user. The sliding window method is used to calculate the rate of change of the optimal estimated target user's physiological parameters, and anomaly detection is performed in combination with the target early warning parameters; If the heart rate exceeds the target range or the blood oxygen saturation falls below the warning threshold for three consecutive samplings, the health risk warning mechanism will be triggered, and a warning message will be pushed to the user's terminal.

[0010] Optionally, the step of performing trend analysis on the target user's physiological parameters and behavioral data during routine monitoring, and generating health intervention suggestions for abnormal fluctuations or potential health risks, includes: In the daily monitoring process, the time series decomposition method is used to monitor the long-term changing trend of the target user's physiological parameters. The time series decomposition method applies a trend term, periodic term and random term separation model. Based on the separation model of trend, period, and random terms, determine whether there are abnormal trends in the physiological parameters of the target user; If no abnormal trend is found, the user's health status is determined to be stable, and routine monitoring continues. If an abnormal trend is detected, it is determined that there is a potential risk to the user's health status, triggering a health intervention suggestion generation program. The program then pushes health intervention suggestions based on the user's exercise behavior data. These suggestions include adjustments to exercise intensity, optimization of daily routines, and dietary recommendations.

[0011] Optionally, the process of acquiring users' historical health data and medical diagnosis results, performing big data analysis on physiological parameters, exercise behavior data, and historical health data, and continuously refining and improving the health status model based on the inherent relationships between the various data points includes: The user's historical health data includes past medical history, medication records, and physical examination reports; It can acquire historical health data and medical diagnosis results authorized by users, and obtain long-term physiological parameters and exercise behavior data from smart wearable devices; The historical health data, medical diagnosis results, physiological parameters, and exercise behavior data are aligned and fused. The isolated forest algorithm is used to identify outlier samples, and interpolation is performed to complete data with missing key features. Based on the fused data, feature engineering methods are used to expand the dimensions and extract time-domain features, frequency-domain features, and nonlinear features. By utilizing the extracted time-domain features, frequency-domain features, and nonlinear features, a transfer learning algorithm is used to update the health status model, and to mine the association rules between the physiological parameters, the exercise behavior data, and the risk of disease occurrence, resulting in the rule that "if the nocturnal heart rate variability decreases and the proportion of deep sleep decreases, then the cardiovascular risk increases." An online learning method is used to revise the health status model, and the parameters of the health status model are updated using incremental gradient descent to adapt to the dynamic changes in the user's health status.

[0012] Optionally, the step of acquiring health status models under different user group characteristics, establishing a health knowledge base and case library for personalized health monitoring, and forming standardized health assessment schemes includes: Based on the health status models under different user group characteristics, monitoring data and model evaluation results of users of different ages, genders and health conditions were obtained, and the K-means algorithm was used to segment users. Based on the user segmentation results, the distribution of health indicators of each group is statistically analyzed, and the optimal warning threshold for each group is determined by ROC curve. Establish a health knowledge base based on rules and cases, which includes diagnostic rules in the form of if-then and typical health cases; A case library is constructed using knowledge graph technology. The case library includes the relationships between physiological parameters, diseases, symptoms and intervention measures. Relevant knowledge is retrieved through semantic similarity calculation. The system receives basic information and initial health data from new users, performs rule matching based on the health knowledge base, retrieves similar cases from the case library, and uses Bayesian inference methods to derive personalized health assessment plans and early warning parameters.

[0013] Optionally, the process of incorporating user health monitoring data, early warning records, and intervention effects into a health management platform for comprehensive management, tracking changes in user health status in real time, assessing key health indicators and potential risks, and generating health reports from the monitoring results includes: The system receives a health monitoring service request carrying a user identifier, retrieves the user's health record and historical monitoring data according to the service request, and generates a personalized monitoring plan based on the health record and the historical monitoring data. The system acquires physiological parameters, exercise data, and warning events sent in real time by smart wearable devices, tracks key health indicators and warning frequency in real time, and determines whether the physiological parameters and exercise data are within the normal range and whether the warning events are handled in a timely manner. If the physiological parameters exceed the normal range, the exercise data is abnormal, or the warning event is not handled in a timely manner, a health reminder will be automatically sent to the user and authorized guardian through the message push engine. Receive health report generation instructions, and automatically generate periodic health reports and annual health assessment reports based on the generation instructions, combined with user health trends, early warning records, and intervention effects.

[0014] Furthermore, to achieve the above objectives, the present invention also provides an artificial intelligence-based intelligent wearable health monitoring system, the artificial intelligence-based intelligent wearable health monitoring system comprising: The data modeling module is used to acquire the user's multimodal physiological parameters and behavioral data through smart wearable devices, determine the user's health indicators based on the multimodal physiological parameters, and establish a nonlinear mapping relationship between the user's health indicators and the behavioral data to obtain a health status model. The parameter optimization module is used to optimize and solve the personalized health warning thresholds for target users based on the constructed health status model, so as to obtain target warning parameters that are suitable for the health status of target users. The real-time early warning module is used to configure the optimized target early warning parameters to the smart wearable device, collect the target user's physiological parameters in real time and perform dynamic monitoring, and trigger the health risk early warning mechanism when the physiological parameters are abnormal. The trend intervention module is used to perform trend analysis on the physiological parameters and behavioral data of target users during daily monitoring, and to generate health intervention suggestions for abnormal fluctuations or potential health risks. The model correction module is used to acquire users' historical health data and medical diagnosis results, perform big data analysis on physiological parameters, exercise behavior data and historical health data, and continuously correct and improve the health status model based on the inherent relationship between the data. The knowledge building module is used to acquire health status models under the characteristics of different user groups, establish a health knowledge base and case library for personalized health monitoring, and form standardized health assessment schemes. The comprehensive management module is used to integrate user health monitoring data, early warning records, and intervention effects into the health management platform for comprehensive management. It tracks changes in user health status in real time, assesses key health indicators and potential risks, and generates health reports from the monitoring results.

[0015] This invention provides an artificial intelligence-based intelligent wearable health monitoring method. The method utilizes nonlinear mapping modeling of multimodal physiological parameters and behavioral data, overcoming the limitations of traditional single-parameter assessments. It comprehensively captures the intrinsic correlation between individual health characteristics and behavioral habits, enabling differentiated health status analysis and improving assessment accuracy. By optimizing the health status model to generate individualized early warning thresholds, it replaces traditional standardized thresholds, fully considering user physiological characteristics, historical health data, and individual differences. This significantly improves the targeting and accuracy of abnormal indicator identification and reduces false alarm / missed alarm rates. Relying on real-time dynamic monitoring and trend analysis, it can not only respond instantly to physiological abnormalities but also target potential... Risks are proactively identified and intervention suggestions are generated, shifting from passive monitoring to proactive health management, enhancing the foresight and effectiveness of user health interventions. Historical health data and medical diagnostic results are integrated, and health status models are dynamically revised through big data analysis, allowing the system to continuously evolve with changes in user health status and adapt to individual health needs in the long term. Simultaneously, a group knowledge base and standardized solutions are constructed to balance personalized monitoring and large-scale health management needs. Through a health management platform, monitoring data, early warning records, and intervention effects are integrated to achieve unified management of health indicator tracking, risk assessment, and report generation, providing comprehensive and real-time health decision support for users and healthcare providers, and promoting the implementation of precision health management. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the structure of an AI-based intelligent wearable health monitoring device in the hardware operating environment of the embodiment of the present invention. Figure 2 This is a flowchart illustrating an embodiment of the intelligent wearable health monitoring method based on artificial intelligence according to the present invention. Figure 3 This is a flowchart illustrating the optimization steps of an artificial intelligence-based intelligent wearable health monitoring method according to an embodiment of the present invention, which optimizes the personalized health warning threshold of the target user based on the constructed health status model and uses a genetic algorithm for intelligent optimization. Figure 4 This is a schematic diagram illustrating the specific steps involved in forming a standardized health assessment scheme according to an embodiment of the artificial intelligence-based intelligent wearable health monitoring method of the present invention. Figure 5 This is a structural block diagram of an embodiment of the intelligent wearable health monitoring system based on artificial intelligence according to the present invention.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] Reference Figure 1 , Figure 1 This is a schematic diagram of the structure of an AI-based smart wearable health monitoring device, which is part of the hardware operating environment of an embodiment of the present invention.

[0020] like Figure 1 As shown, the AI-based smart wearable health monitoring device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen, and optionally, it may also include a standard wired interface or a wireless interface. In this invention, the wired interface of the user interface 1003 may be a USB interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be a high-speed random access memory (RAM) or a non-volatile memory (NVM), such as a disk storage device. The memory 1005 may also optionally be a storage device independent of the aforementioned processor 1001.

[0021] Those skilled in the art will understand that Figure 1 The structure shown does not constitute a limitation on AI-based smart wearable health monitoring devices, which may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0022] like Figure 1 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an artificial intelligence-based smart wearable health monitoring program.

[0023] exist Figure 1 In the AI-based smart wearable health monitoring device shown, the network interface 1004 is mainly used to connect to the backend server and communicate data with the backend server; the user interface 1003 is mainly used to connect to peripherals; the AI-based smart wearable health monitoring device calls the AI-based smart wearable health monitoring program stored in the memory 1005 through the processor 1001 and executes the AI-based smart wearable health monitoring method provided in the embodiment of the present invention.

[0024] Based on the above hardware structure, an embodiment of the intelligent wearable health monitoring method based on artificial intelligence of the present invention is proposed.

[0025] Reference Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the intelligent wearable health monitoring method based on artificial intelligence according to the present invention.

[0026] In one embodiment, the AI-based smart wearable health monitoring method includes the following steps: Step S100: Obtain the user's multimodal physiological parameters and behavioral data through a smart wearable device, determine the user's health indicators based on the multimodal physiological parameters, and establish a nonlinear mapping relationship between the user's health indicators and behavioral data to obtain a health status model.

[0027] Multimodal physiological parameters refer to various physiological signals collected through wearable devices, including but not limited to heart rate, blood oxygen saturation, and skin conductance, used to comprehensively reflect the user's health status. These parameters are continuously collected by sensors built into the wearable device, such as photoplethysmography (PPG) sensors and electrode pads. Behavioral data refers to the user's daily activity records, including exercise volume, sleep quality, and resting time, used to analyze the impact of lifestyle habits on health. This data can be obtained through motion sensors and accelerometers on the wearable device or through user input. Nonlinear mapping relationships refer to complex associations between parameters established through mathematical or machine learning models. For example, recurrent neural networks (RNNs) or graph convolutional networks (GCNs) are used to capture the dynamic interactions between multidimensional data to quantify the interactive influence of physiological parameters and behavioral data on health status. The health status model is a mathematical framework built based on the above data. Its functions include quantifying the influence weight of each parameter on health status, such as using random forests or support vector machines (SVMs) to achieve classification or regression prediction. The technical operation includes data acquisition, feature extraction, and model training: After wearable device sensors continuously collect multimodal data, noise is eliminated through filtering and standardization. The processed data is then input into a pre-trained model for feature fusion, and finally, the model parameters are optimized using a backpropagation algorithm to minimize prediction error. This process, through multi-dimensional data correlation, overcomes the limitations of traditional single-parameter assessments and improves the comprehensiveness of health status analysis.

[0028] Step S200: Based on the constructed health status model, optimize and solve the personalized health warning threshold for the target user to obtain the target warning parameters applicable to the target user's health status.

[0029] Personalized health alert thresholds refer to critical values ​​for specific user health indicators. For example, they dynamically adjust the normal range of resting heart rate based on user age, basal metabolic rate, and other characteristics, and can be calculated using historical health data or biometric information. Optimization involves finding the optimal combination of thresholds under the constraints of a health status model using mathematical optimization algorithms, such as gradient descent or genetic algorithms. The technical process includes objective function construction, constraint setting, and iterative optimization: First, an objective function maximizing alert accuracy is defined, while the fluctuation range of historical user data and differences in physiological characteristics are used as constraints. Through multiple iterative calculations, threshold parameters that balance the false alarm rate and the false negative rate are determined. This process, by dynamically adjusting the thresholds, eliminates the universality defects of traditional standardized thresholds and improves the accuracy of abnormal indicator identification.

[0030] Step S300: Configure the optimized target warning parameters to the smart wearable device, collect the target user's physiological parameters in real time and perform dynamic monitoring, and trigger the health risk warning mechanism when the physiological parameters are abnormal.

[0031] Dynamic monitoring refers to wearable devices continuously collecting physiological parameters and comparing them in real time to warning thresholds. For example, heart rate data is collected once per second and compared with a preset threshold range, achieved through a built-in microcontroller or edge computing module. The health risk warning mechanism includes threshold exceedance detection, risk grading, and alarm triggering: when monitored values ​​exceed the threshold range, the system initiates different levels of warnings based on the degree of deviation, such as minor abnormality alerts or emergency alarms, and notifies the user or caregiver via vibration, application push notifications, etc. This process shortens the risk detection time window through immediate response and improves the real-time handling capability of abnormal states.

[0032] Step S400: During routine monitoring, trend analysis is performed on the physiological parameters and behavioral data of the target users, and health intervention suggestions are generated for abnormal fluctuations or potential health risks.

[0033] Trend analysis refers to time-series analysis of long-term monitoring data. For example, calculating the 7-day average trend of heart rate variability (HRV) using the sliding window method can be achieved through statistical methods or time-series models (such as ARIMA). Health intervention recommendation generation involves rule engines or machine learning models. For instance, when a persistent trend of elevated blood pressure is detected, the system calls association rules from a knowledge base to recommend a low-salt diet. The technical operations include data serialization, pattern recognition, and recommendation matching: after converting multi-dimensional data into a time-series format, a Long Short-Term Memory (LSTM) network is used to identify potential risk patterns. Finally, a pre-set or adaptive intervention strategy is matched based on the risk type. This process reduces the probability of disease occurrence through proactive intervention strategies, achieving proactive health management.

[0034] Step S500: Obtain the user's historical health data and medical diagnosis results, perform big data analysis on physiological parameters, exercise behavior data and historical health data, and continuously revise and improve the health status model based on the inherent relationship between the data.

[0035] Big data analytics refers to integrating multi-source heterogeneous data for feature engineering and pattern mining. This multi-source heterogeneous data includes medical diagnostic reports and genetic testing results. For example, association rule algorithms can be used to discover statistical associations between specific exercise patterns and blood glucose fluctuations. Model refinement includes two methods: online learning and transfer learning. Online learning continuously fine-tunes model parameters through real-time data streams, while transfer learning transfers common patterns from group data to individual models. Technical operations include data cleaning, feature fusion, and model retraining. First, missing values ​​in medical reports are eliminated. Then, structured medical data is aligned with wearable data. Finally, cross-validation is used to evaluate the model update effect. This process enables the health status model to evolve, adapting to long-term physiological changes in users and improving predictive accuracy.

[0036] Step S600: Obtain health status models under the characteristics of different user groups, establish a health knowledge base and case library for personalized health monitoring, and form a standardized health assessment plan.

[0037] A health knowledge base is a structured database that stores the health characteristics and intervention effects of a population. For example, it records typical risk patterns and effective interventions for users of different age groups. Clustering algorithms (such as K-means) can be used to group user groups and extract common features to construct a knowledge graph. Standardized solutions refer to general assessment processes derived from the knowledge base, such as developing standardized data collection protocols for diabetes risk screening. The technical operations include steps such as feature clustering, pattern mining, and solution template generation: after constructing a knowledge graph by extracting common features from each group, validated effective intervention strategies are transformed into configurable standardized templates. This process balances the needs of personalization and scalability, supporting individualized monitoring while facilitating large-scale health management implementation.

[0038] Step S700: Integrate user health monitoring data, early warning records, and intervention effects into the health management platform for comprehensive management, track changes in user health status in real time, assess key health indicators and potential risks, and generate health reports from the monitoring results.

[0039] A health management platform is a software system integrating data storage, analysis, and interaction. It includes modules such as a data lake storing raw monitoring data, a real-time computing engine processing alert requests, and a visual interface displaying health trends. Health report generation involves data aggregation and intelligent summarization; for example, natural language processing (NLP) technology transforms structured data such as abnormal indicators and intervention suggestions into graphic reports. Technical operations include data pipeline construction, multi-dimensional analysis, and report template configuration: integrating data sources such as wearable devices and hospital systems through application programming interfaces (APIs), using principal component analysis (PCA) to reduce dimensionality and display changes in key indicators, and finally customizing report content according to user needs. This process forms a complete health management loop, providing users with traceable health records while assisting healthcare professionals in making accurate decisions.

[0040] This embodiment provides an AI-based intelligent wearable health monitoring method that enhances the comprehensiveness and dynamic adaptability of health assessment through multimodal data fusion and nonlinear modeling, mitigates false alarms and missed alarms through personalized early warning threshold design, reduces the probability of potential risks through trend analysis and proactive intervention strategies, enhances the system's evolutionary capabilities through continuous model optimization and knowledge base construction, and achieves full-process integration of monitoring, early warning, intervention, and assessment through platform-based management. This method overcomes the shortcomings of traditional methods in terms of dynamism, personalization, and long-term sustainability, resulting in a more efficient and intelligent health monitoring and management system.

[0041] In one embodiment, multimodal physiological parameters include heart rate, blood oxygen saturation, and skin temperature; behavioral data includes steps, exercise type, exercise intensity, and sleep structure; the user's multimodal physiological parameters and behavioral data are acquired through a smart wearable device, user health indicators are determined based on the multimodal physiological parameters, and a nonlinear mapping relationship between user health indicators and behavioral data is established to obtain a health status model, including: The system acquires users' multimodal physiological parameters and behavioral data through smart wearable devices, and preprocesses the multimodal physiological parameters and behavioral data, including data cleaning, normalization and feature extraction. Multimodal physiological parameters can be quantitative indicators reflecting the physiological function of the human body, which can be obtained in real time through biosensors. Examples include heart rate, blood oxygen saturation, and skin temperature. Heart rate is measured in beats per minute, blood oxygen saturation is expressed as a percentage of the proportion of oxyhemoglobin in the blood, and skin temperature is measured in degrees Celsius. Behavioral data can be structured records describing the user's daily activity patterns, which can be obtained through motion sensors and sleep monitoring algorithms. Examples include steps, exercise type, exercise intensity, and sleep structure. Steps are counted using an accelerometer, exercise types are categorized with tags such as brisk walking, running, and cycling, exercise intensity is divided into low, medium, and high levels based on exercise energy consumption, and sleep structure includes time-series distribution data such as deep sleep duration, light sleep duration, and REM cycle ratio. Data cleaning can be a process of removing sensor noise or filling in missing values. Examples include using wavelet transform to eliminate motion artifacts in heart rate monitoring or using linear interpolation to fill in data gaps caused by temporary device disconnections. Normalization can be an operation that converts parameters with different dimensions to a uniform scale. For example, heart rate (range 60 to 120 bpm) and skin temperature (range 2836°C) can be mapped to the [0,1] interval. Feature extraction can be a process of extracting key statistical features through time-domain or frequency-domain analysis. For example, heart rate variability (HRV) indicators such as standard deviation (SDNN), standard deviation of blood oxygen saturation (SD_SpO2), and derived features such as the proportion of deep sleep time in sleep structure can be calculated. Through the above preprocessing process, data noise interference can be reduced, the comparability of cross-user data can be enhanced, and the dimensionality of model input can be reduced.

[0042] Based on the preprocessed multimodal physiological parameters, user health indicators were determined, including cardiovascular health score, metabolic health score and sleep health score; The cardiovascular health score can be a comprehensive indicator quantifying cardiac function and autonomic nervous system regulation, for example, calculated using a weighted combination of resting heart rate and heart rate variability (HRV). The metabolic health score can be a quantitative parameter assessing energy metabolism efficiency and basal metabolic rate, for example, weighted summation combining skin temperature change trends and calorie expenditure data. The sleep health score can be a comprehensive evaluation reflecting sleep quality and structural integrity, for example, calculated based on a weighted average of the percentage of deep sleep duration, REM sleep cycle proportion, and the standard deviation of light sleep duration. By breaking down multimodal physiological parameters into sub-dimension scores such as cardiovascular, metabolic, and sleep scores, health status assessments can be made more targeted, for example, distinguishing between the improvement in metabolic efficiency caused by high-intensity exercise and its potential interference with short-term sleep structure.

[0043] A health status model is obtained by establishing a nonlinear mapping relationship between user health indicators and behavioral data through a long short-term memory network algorithm. The Long Short-Term Memory (LSTM) network algorithm can be a recurrent neural network architecture with a gating mechanism. For example, through the synergistic effect of input gates, forget gates, and output gates, it captures the temporal dependencies between behavioral data and health indicators. Behavioral data includes patterns of exercise intensity over time, while health indicators include long-term trends in metabolic scores. Specifically, it involves constructing input-output pairs from preprocessed behavioral data (such as daily exercise intensity sequences) and corresponding health indicators (such as metabolic scores over the past week). Backpropagation is then used to optimize network parameters, enabling the model to learn the cumulative impact of long-term behavioral patterns on health status. Compared to traditional fully connected networks, this method retains the influence of historical behavioral data through memory units, such as the dynamic process of long-term regular exercise gradually improving cardiovascular scores, thereby enhancing the model's accuracy in predicting the evolution of health status.

[0044] This embodiment defines the specific types of multimodal physiological parameters and the subdivision dimensions of behavioral data. It combines a standardized preprocessing process to eliminate noise and dimensional differences, further decomposes health status into sub-scores such as cardiovascular, metabolic, and sleep to achieve classification modeling, and uses long short-term memory networks to capture the temporal correlation between behavioral patterns and health indicators. This can improve the accuracy and dynamic adaptability of health status assessment, enhance the interpretability of the model for long-term behavioral intervention effects, and ultimately provide structured data support for personalized health early warning and intervention strategies.

[0045] In one embodiment, reference Figure 3 The health warning thresholds include the resting heart rate abnormality threshold, the upper limit of the active heart rate, and the lower limit of blood oxygen saturation. The personalized health warning thresholds for the target user are optimized based on the constructed health status model. The optimization is carried out using a genetic algorithm intelligent optimization algorithm, which includes the following steps.

[0046] S201: Based on the preset chromosome encoding method, the three physiological parameters of abnormal resting heart rate threshold, upper limit of active heart rate and lower limit of blood oxygen saturation are encoded into binary strings, and the length of the binary strings is set according to the parameter accuracy requirements. Chromosome encoding can be used to map the parameters to be optimized into an algorithm-processable encoding format. The number of binary bits is determined by the parameter value range and precision requirements. For example, when the resting heart rate threshold range is set to [50, 100] bpm and the precision is 1 bpm, a 7-bit binary encoding is required. The binary string can be a specific representation of the chromosome encoding method. For example, a resting heart rate threshold of 75 bpm can be encoded as the binary sequence "01001011". The parameter precision requirement can be the minimum step size for threshold adjustment, determined by the mathematical relationship between the number of binary bits and the parameter value range. For example, when the heart rate threshold needs to be accurate to 1 bpm, the number of binary bits must cover the target range. Parameter encoding and precision setting are achieved by discretizing continuous parameters into binary form. For example, three physiological parameters are converted into independent binary strings and their corresponding number of bits are set, thereby ensuring that the minimum step size for threshold adjustment meets clinical needs.

[0047] S202: Obtain the predefined binary string, set the fitness function of the genetic algorithm, and use the normalization method to convert the behavioral data into a dimensionless index; The fitness function of the genetic algorithm can be a mathematical expression evaluating the merits of an individual. It is defined by integrating the correlation between physiological parameters and behavioral data. For example, the fitness function can be designed as (1 - health risk probability) × normalized behavioral data value. The normalization method can be a numerical conversion technique to eliminate interference from different units of measurement. For example, the daily step count range [0, 20000] can be mapped to the interval [0, 1]. Behavioral data can be records of the user's daily activities, such as exercise volume and sleep duration, obtained through sensors or user input. The fitness function is constructed by combining the normalized behavioral data with the risk probability output by the health status model. For example, the formula fitness = (1 - risk probability) × normalized behavioral value can be used, thereby achieving a comprehensive evaluation of the applicability of threshold combinations to individual users.

[0048] S203: After the fitness function is set, a roulette wheel selection operation is used to select excellent individuals, and the crossover probability of the roulette wheel selection operation is set to a preset value. Gene recombination is carried out using a single-point crossover method. The roulette wheel selection operation can be a selection mechanism based on the fitness probability of individuals, simulating the natural selection process through random selection. For example, individuals with higher fitness have a higher probability of being selected. The crossover probability can be a parameter that controls the frequency of crossover operations; for example, it can be set to 0.8 to balance the algorithm's exploration and development capabilities. The single-point crossover method can be a recombination method in which gene segments are exchanged at random positions on parent chromosomes. For example, two binary strings "A=1010" and "B=1100" will generate "A'=1110" and "B'=1000" after the second crossover. Genetic operations and parameter configuration are achieved by setting the crossover probability and performing roulette wheel selection and single-point crossover. For example, setting the crossover probability to 0.8 and then performing gene recombination on the selected parent individuals can achieve the technical effect of introducing diversity while maintaining parent characteristics and preventing local optima.

[0049] S204: Based on the selected superior individuals, iteratively run the genetic algorithm, and set the population size of the genetic algorithm to the first preset value and the maximum number of iterations to the second preset value; The population size can be the number of individuals the algorithm processes simultaneously; for example, 100 individuals can be set to balance computational efficiency and search scope. The maximum number of iterations can be one of the conditions for algorithm termination; for example, 200 iterations can be set to control computational resource consumption. Iterative execution can be the process by which the algorithm optimizes itself through operations such as selection, crossover, and mutation; for example, a new population is generated and the optimal solution is updated in each iteration. Algorithm parameter configuration is achieved by setting the population size and the number of iterations. For example, setting the population size to 100 and the maximum number of iterations to 200 before starting the algorithm loop can achieve the technical effect of controlling computational overhead while ensuring optimization accuracy.

[0050] S205: When the genetic algorithm reaches the convergence condition or the maximum number of iterations, decode to obtain the personalized health warning threshold for the target user, and use it as the target warning parameter applicable to the target user's health status.

[0051] The convergence condition can be a termination criterion where the change in the fitness function value is less than a preset threshold. For example, the computation stops when the fitness change is less than 0.01 for 10 consecutive iterations. Decoding can be the process of converting a binary string into specific physiological parameter values. For example, the binary string "01001011" is decoded into a resting heart rate threshold of 75 bpm. The target warning parameter can be a personalized combination of thresholds obtained from the final optimization. For example, it can include a resting heart rate of 75 bpm, an upper limit of active heart rate of 140 bpm, and a lower limit of blood oxygen saturation of 90%. Convergence and decoding are achieved by monitoring changes in fitness and performing binary-to-decimal conversion. For example, when the number of iterations reaches 200 or fitness converges, the binary string of the optimal individual is converted into a specific value, thereby achieving the technical effect of breaking through local optima and obtaining the global optimal solution.

[0052] The fitness function formula for the personalized health warning threshold for the target user can be: ; In the formula, For the fitness function, This is the threshold for abnormal resting heart rate. This is the upper limit of the active heart rate. This is the lower limit of blood oxygen saturation. The weight of the k-th health indicator satisfies , Based on the current threshold The output value of the health status model. These are the actual measured health indicator values. This represents the theoretical maximum value of the k-th health indicator. This represents the theoretical minimum value of the k-th health indicator.

[0053] It should be noted that the fitness function quantifies the health warning threshold (R, A, S) by the reciprocal of the normalized weighted root mean square error. The smaller the error, the closer the model output is to the actual data, and the higher the fitness value, which aligns with the optimization objective of maximizing fitness in genetic algorithms. The reciprocal relationship transforms minimizing error into maximizing fitness, facilitating the selection operation of the genetic algorithm; that is, individuals with high fitness are more likely to be selected. It covers three core health indicators: the abnormal resting heart rate threshold, the upper limit of active heart rate, and the lower limit of blood oxygen saturation. For example, when k=1, it refers to indicators related to the abnormal resting heart rate threshold, such as resting heart rate stability and abnormal frequency; when k=3, it refers to indicators related to the lower limit of blood oxygen saturation, such as blood oxygen supply and demand balance and tissue oxygenation status. It reflects the different importance of various indicators in health assessment. For example, in clinical practice, the urgency of blood oxygen saturation may be higher than that of resting heart rate. By pre-setting or learning to adjust the weights, it can be adapted to different populations. For example, athletes focus on heart rate indicators, while patients with respiratory diseases focus on blood oxygen indicators. It directly measures the deviation between model predictions and actual data; the smaller the error, the closer the threshold is to the user's health status. This approach eliminates the differences in the dimensions of different indicators, allowing for cross-indicator comparison of errors and converting absolute errors into relative errors, thus preventing large numerical indicators from dominating error calculations. It should be understood that the fitness function becomes the core evaluation criterion of the genetic algorithm, directly guiding the direction of threshold optimization and ensuring that the calculated threshold conforms to both physiological logic and the individual user's data characteristics.

[0054] This embodiment achieves discretization by encoding physiological parameters into binary strings and setting precision requirements. After normalization to eliminate differences in the dimensionality of behavioral data, a fitness function is constructed. Iterative optimization of the population is achieved by combining roulette wheel selection and single-point crossover. The population size and number of iterations are set to control computational efficiency. Finally, the globally optimal personalized threshold combination is obtained through decoding. This can achieve the technical effects of improving early warning accuracy through multi-parameter collaborative optimization, enhancing individual adaptability by fusing behavioral data, and reducing local optimum bias by utilizing the global search capability of genetic algorithms.

[0055] In one embodiment, after the genetic algorithm reaches the convergence condition or the maximum number of iterations, decodes and obtains the personalized health warning threshold for the target user, and uses it as a target warning parameter applicable to the target user's health condition, the method further includes: Multimodal physiological parameters are encoded as parameter combinations for individuals, and an initial population is randomly generated. In this process, the parameter combination encoding for an individual can be achieved by combining multiple physiological parameters into a parameter vector and mapping it to an individual in the genetic algorithm. Chromosome representation can be implemented using binary or floating-point numbers. For example, three parameters can be encoded as binary strings and concatenated to form an individual chromosome. Randomly generating the initial population can be done using random algorithms such as uniform distribution to generate a set of parameter combinations. For example, the resting heart rate threshold is randomly selected within the range [50, 100], and the upper limit of the active heart rate is randomly generated within the range [100, 180]. Parameter combination encoding and initial population generation can be implemented using random algorithms, thereby ensuring the diversity of the parameter space covered by the initial population and preventing the algorithm from getting trapped in local optima due to initial solution selection bias.

[0056] The fitness of each individual is calculated based on the health status model, and superior individuals are selected and new offspring individuals are generated through crossover mutation. The health status model can be an assessment system that integrates historical user data, behavioral data, and model-predicted risk. Behavioral data matching can be an indicator of the adaptability of parameter combinations to users' daily activity patterns; for example, active users may require a higher upper limit for their active heart rate. Crossover and mutation operations can involve exchanging chromosome segments from parent individuals at a single point and randomly flipping chromosome positions with a preset probability; for example, segments of the resting heart rate threshold and blood oxygen threshold can be exchanged. Fitness calculation and individual selection can dynamically evaluate the comprehensive performance of parameter combinations through the health status model. Crossover and mutation operations explore the parameter space through recombination and random perturbation, thereby selecting parameter combinations that both reduce health risks and conform to user behavioral characteristics.

[0057] Through multiple iterations of evolutionary operations and optimization of parameter combinations, the individual with the highest fitness is finally selected, and its corresponding parameter combination is used as the target warning parameter.

[0058] Iterative evolution can involve repeatedly executing selection, crossover, and mutation processes until a termination condition is met. For example, termination conditions may include a preset number of iterations or convergence of fitness. The final individual selection can be based on a comparison of fitness values ​​after multiple generations of iterations; for example, the individual with the highest fitness is selected as the final solution. Iterative evolution and optimal solution selection gradually approach the global optimum in the parameter space through multiple generations of iteration, thereby achieving a balance between individual fitness and the accuracy of risk prediction.

[0059] This embodiment achieves collaborative optimization of parameter combinations by encoding multimodal physiological parameters into individuals, ensuring the diversity of the parameter space by randomly generating the initial population, dynamically evaluating the comprehensive performance of parameter combinations through a health status model, maintaining the exploratory ability of the population evolution process through crossover and mutation operations, and balancing algorithm efficiency and the approximation accuracy of the global optimum through an iterative mechanism. This can achieve the technical effects of improving the personalized accuracy of health warning thresholds, enhancing algorithm robustness, and reducing the risk of local optima. At the same time, through the dynamic adaptation of parameter combinations and user behavior data, reliable warning parameter combinations that conform to the physiological characteristics and lifestyles of different users can be provided.

[0060] In one embodiment, the optimized target warning parameters are configured into a smart wearable device to collect and dynamically monitor the target user's physiological parameters in real time. A health risk warning mechanism is triggered when the physiological parameters are abnormal, including: The optimized target warning parameters are obtained and sent to the local processor of the smart wearable device via Bluetooth Low Energy protocol. Bluetooth Low Energy (BLE) is a wireless communication standard designed for low-power devices, enabling efficient data transfer between smart wearable devices and external systems. Its core advantage lies in balancing transmission speed and power consumption, making it suitable for scenarios requiring long-term continuous operation of wearable devices. Target warning parameters can be generated through optimized algorithms, such as personalized thresholds, and data security is ensured through encrypted communication. For example, this protocol can achieve stable transmission of target warning parameters, preventing threshold invalidation due to communication interruptions.

[0061] Multi-sensor fusion technology is used to continuously collect users' heart rate and blood oxygen saturation data, and particle filtering algorithm is used to obtain the optimal estimated physiological parameters of the target user. Multi-sensor fusion technology integrates redundant or complementary data from multiple sensors (such as heart rate and blood oxygen sensors) and uses algorithms (such as particle filtering) to eliminate noise and improve the accuracy of parameter estimation. For example, it can combine accelerometer data to correct heart rate measurement errors caused by motion interference. Particle filtering is a recursive Bayesian filtering method suitable for state estimation of nonlinear, non-Gaussian systems. By generating particle swarms to simulate parameter distributions, it dynamically adjusts weights to track changes in real physiological parameters, outperforming traditional Kalman filtering, especially when dealing with sensor noise or abrupt signals. For instance, this technology can fuse raw signals from photoplethysmography (PPG) sensors and accelerometers to output more accurate estimates of heart rate and blood oxygen saturation.

[0062] The sliding window method is used to calculate the rate of change of the optimal estimated target user's physiological parameters, and anomaly detection is performed in combination with the target early warning parameters; The sliding window method uses a fixed-length time window (e.g., 5 minutes) to perform statistical analysis on the data within the window, such as calculating the mean, standard deviation, or rate of change, to capture short-term trends in physiological parameters. For example, it calculates the rate of change of heart rate per minute to identify abnormal acceleration or deceleration. This method combines a threshold for a target warning parameter, compares the deviation of the current parameter value from the threshold, and analyzes the trend of the rate of change to determine whether there is a risk of continued deviation. For instance, if the heart rate rises more than a preset threshold per minute within the window, it is considered an abnormal trend.

[0063] If the heart rate exceeds the target range or the blood oxygen saturation falls below the warning threshold for three consecutive samplings, the health risk warning mechanism will be triggered, and a warning message will be pushed to the user's terminal.

[0064] This trigger condition determines a high-risk state by setting a consecutive threshold exceeding a certain number of times (e.g., three times). For example, if the system detects that the heart rate consistently exceeds the personalized upper limit (e.g., 120 beats / minute) or the blood oxygen saturation falls below the warning lower limit (e.g., 85%) for three consecutive samplings, it will determine a high-risk state and trigger an alert. This design reduces the risk of false alarms caused by transient noise or brief physiological fluctuations by requiring multiple consecutive threshold exceedances. For instance, this mechanism can push alert information to the user's mobile phone or emergency contact terminal to ensure timely response.

[0065] This embodiment achieves efficient configuration of target warning parameters through Bluetooth Low Energy protocol, improves the accuracy of physiological parameter estimation by utilizing multi-sensor fusion and particle filtering algorithms, enhances the reliability of anomaly detection by combining sliding window method and continuous judgment strategy, and finally reduces the false alarm rate by using multi-level threshold breakthrough conditions. These technical means work together to effectively suppress noise from single sensors or environmental interference, capture gradual physiological change trends, and ensure long-term stable operation and real-time response capability of the device, significantly improving the accuracy, reliability, and practicality of health risk warning.

[0066] In one embodiment, during routine monitoring, trend analysis is performed on the target user's physiological parameters and behavioral data, and health intervention recommendations are generated for abnormal fluctuations or potential health risks, including: In the daily monitoring process, the time series decomposition method is used to monitor the long-term changing trend of the target user's physiological parameters. The time series decomposition method applies a model that separates the trend term, periodic term, and random term. Time series decomposition is an analytical method that breaks down time series data into trend, periodic, and random components. Its purpose is to separate different components in the data to accurately identify abnormal deviations from long-term trends. For example, time series decomposition can be implemented using STL or classical time series decomposition models, such as using moving averages to extract the trend component, Fourier transforms to analyze periodic fluctuations, and the remaining portion as the random component.

[0067] Based on the separation model of trend, periodic and random terms, determine whether there are abnormal trends in the physiological parameters of the target user; A model for separating trend, periodic, and random components can be a model that decomposes the original physiological parameter sequence into three independent components using mathematical algorithms. The methods for obtaining these components include, but are not limited to, mathematical operations or algorithmic processing. For example, the separation model can be implemented based on moving averages, Fourier transforms, or other decomposition algorithms to distinguish between long-term trends, periodic fluctuations, and random noise.

[0068] If no abnormal trend is found, the user's health status is determined to be stable, and routine monitoring continues. Routine monitoring can be a monitoring mode that continuously collects users' physiological parameters and repeats the above analysis process. Its purpose is to maintain long-term tracking of users' health status.

[0069] If an abnormal trend is detected, it is determined that there is a potential risk to the user's health status, triggering a health intervention suggestion generation program. The program then pushes health intervention suggestions based on the user's exercise behavior data, including adjustments to exercise intensity, optimization of daily routines, and dietary recommendations.

[0070] The health intervention suggestion generation program can be an automated decision-making system based on preset rules or a knowledge base. Its function is to match intervention strategies according to the type of abnormal parameters. For example, the program can generate personalized suggestions by calling a rule base for adjusting exercise intensity, a template for optimizing daily routines, or a library of dietary recommendation plans, combined with user behavior data (such as exercise volume and sleep duration).

[0071] In terms of technical operation, the process is implemented through four core steps: data decomposition, anomaly detection, decision branching, and suggestion generation. Specifically, after the physiological parameters collected by wearable devices are separated into components using a time-series decomposition model, statistical analysis is performed on trend items, such as calculating whether the slope or fluctuation amplitude exceeds the confidence interval of the individual's historical data. If the trend items do not show any anomalies, routine monitoring is maintained; if the trend items show unexpected and continuous changes, such as a resting heart rate increasing by 5 beats / minute per week for two consecutive weeks, an intervention suggestion generation program is triggered in conjunction with behavioral data (such as a sudden decrease in exercise or insufficient sleep). This process includes sub-steps such as data standardization, component decomposition, statistical threshold setting, multi-dimensional correlation analysis, and suggestion template calling. For example, anomalies are detected using Z-score tests or isolated forest algorithms, while strategies such as recommending low-sodium diet plans or adjusting exercise intensity thresholds are used to improve the targeting of suggestions.

[0072] This embodiment employs a time-series decomposition method to separate long-term trends, periodic fluctuations, and random noise in physiological parameters to accurately identify abnormal trends. It then combines this with user behavior data to generate personalized health intervention suggestions. Furthermore, statistical testing or machine learning algorithms are used to improve the accuracy of anomaly detection, and a pre-defined rule base is utilized to dynamically adapt intervention strategies. This achieves the following technical effects: by separating periodic and random interference factors, it reduces the interference of seasonal fluctuations or noise on long-term trend judgment, thereby improving the scientific rigor of health risk prediction and controlling false alarm rates; by combining user behavior data, it enhances the targeting of intervention suggestions, such as distinguishing between excessive exercise and heart rate abnormalities caused by potential cardiovascular problems, avoiding conflicts between suggestions and users' actual behaviors; and by using long-term trend analysis to proactively identify sub-health states, providing intervention measures before symptoms appear, reducing the probability of disease occurrence, and promoting the transformation of health management from passive monitoring to proactive prevention. In addition, this method, through structured decomposition and knowledge-based suggestion generation, ensures that intervention measures conform to medical guidelines and adapt to individual differences, thereby improving user compliance and strengthening the accuracy, operability, and preventative nature of the health management system.

[0073] In one embodiment, the user's historical health data and medical diagnosis results are acquired, and big data analysis is performed on physiological parameters, exercise behavior data, and historical health data. The health status model is then continuously revised and improved based on the inherent relationships between the various data points, including: The user's historical health data includes past medical history, medication records, and physical examination reports; User historical health data can be a multi-dimensional collection of information recording a user's past health status. It can be obtained through medical system interfaces or by user-initiated uploads. For example, past medical history may include records of chronic diseases such as hypertension and diabetes, medication records may include drug names, dosages and treatment cycles, and physical examination reports may include blood biochemical indicators, imaging examination results, etc.

[0074] It can acquire historical health data and medical diagnosis results authorized by users, and obtain long-term physiological parameters and exercise behavior data from smart wearable devices; Historical health data and medical diagnosis results can be accessed through user-authorized electronic health record systems or medical institution interfaces, while physiological parameters and exercise behavior data can be collected in real time and stored long-term through sensors on smart wearable devices.

[0075] Data alignment and fusion were performed on historical health data, medical diagnosis results, physiological parameters and exercise behavior data. The isolated forest algorithm was used to identify outliers and interpolation was used to complete data with missing key features. Data alignment refers to matching the spatiotemporal information of multi-source data (such as hospital reports and wearable device data) using dimensions such as unified timestamps and user identifiers. For example, hash algorithms can be used to standardize and map user identities. The Isolation Forest algorithm is a tree-structure-based anomaly detection method that separates anomalous samples (such as abnormally high blood pressure values ​​or contradictory medication records) from normal samples by randomly partitioning the data space. Missing key features refer to samples lacking parameters crucial for health assessment (such as missing heart rate variability data). Interpolation completion can be achieved through linear interpolation, time series prediction models (such as ARIMA), or K-nearest neighbor interpolation based on nearby samples. For example, forward imputation can be used to complete continuously missing heart rate data.

[0076] Based on the fused data, feature engineering methods are used to expand the dimensions and extract time-domain features, frequency-domain features, and nonlinear features. Time-domain features can be directly based on time-series statistics, such as the mean, standard deviation, or trend slope of heart rate; frequency-domain features can be extracted through Fourier transform or wavelet analysis, such as the high-frequency power ratio of HRV signals; nonlinear features can be indicators used to quantify signal complexity, such as fractal dimension and Lyapunov exponent. Dimensionality expansion is achieved through signal segmentation, transform domain conversion, and dynamic analysis. For example, after segmenting the PPG waveform, the time-domain statistics of each segment are calculated, and the frequency components are extracted by converting to the frequency domain using FFT.

[0077] By utilizing the extracted time-domain features, frequency-domain features, and nonlinear features, a transfer learning algorithm is used to update the health status model, and to mine the association rules between the physiological parameters, the exercise behavior data, and the risk of disease occurrence, resulting in the rule that "if the nocturnal heart rate variability decreases and the proportion of deep sleep decreases, then the cardiovascular risk increases." Transfer learning algorithms can transfer knowledge from pre-trained models to individual user data. For example, pre-trained models, such as diabetes risk prediction models built on population data, reduce the amount of individual data required through feature mapping and parameter fine-tuning. Association rule mining searches for high-confidence patterns in the feature space using frequent itemset algorithms (such as Apriori or FPGrowth), for example, discovering a statistical association between "decreased nocturnal heart rate variability and reduced proportion of deep sleep" and "increased cardiovascular risk."

[0078] An online learning method is used to revise the health status model, and the parameters of the health status model are updated using incremental gradient descent to adapt to the dynamic changes in the user's health status.

[0079] Online learning continuously updates model parameters through real-time data streams. Incremental gradient descent adjusts model weights based solely on the gradient information of each new data sample received, such as the connection weights of a neural network. For example, when a user continuously uploads sleep quality data for multiple days, the model parameters will be gradually adjusted to reflect changes in cardiovascular risk thresholds.

[0080] This embodiment acquires and fuses multi-source health data, improves data quality using the isolated forest algorithm and interpolation completion technology, expands the dimensions of data representation through the joint extraction of time-domain, frequency-domain, and nonlinear features, dynamically updates model parameters using transfer learning and online learning strategies, and enhances the interpretability of risk prediction through association rule mining. The following technical effects can be achieved: Anomaly detection and data completion techniques reduce the interference of noisy data on model training, ensuring the reliability of input data; multi-dimensional feature engineering captures the statistical regularities, periodic characteristics, and complex dynamic patterns of physiological signals, improving the model's ability to identify complex health risks; transfer learning reduces the amount of individual data required, while online learning allows model parameters to be continuously optimized according to the user's long-term physiological changes, such as adjusting the risk prediction threshold based on blood pressure trends; and explicit association rules (such as the association between sleep and cardiovascular risk) provide interpretable evidence of health risks, enhancing user acceptance of health management recommendations.

[0081] In one embodiment, reference Figure 4 To obtain health status models under the characteristics of different user groups, establish a health knowledge base and case library for personalized health monitoring, and form standardized health assessment programs, including: S601: Based on the health status model under the characteristics of different user groups, obtain the monitoring data and model evaluation results of users of different ages, genders and health conditions, and use the Kmeans algorithm to segment users; The health status model can be a predictive model built based on users' physiological parameters, lifestyle habits, and past medical history. It can be generated through machine learning methods or medical guidelines. For example, model evaluation results may include health indicators such as cardiovascular risk level and probability of metabolic syndrome. The K-means algorithm can be an unsupervised clustering algorithm that iteratively divides multidimensional data into K non-overlapping clusters. The centroid of each cluster represents the typical characteristics of that group. For example, input data may include parameters such as age, heart rate variability coefficient, and sleep quality score. In a specific embodiment, by initializing the K value (e.g., setting 3 groups), the distance between each data point and the centroid is calculated, and cluster affiliation is redistributed until the algorithm converges, thereby identifying user groups with similar health characteristics. This technique uses structured user data to provide a clustering basis for subsequent statistical analysis and threshold optimization.

[0082] S602: Statistical analysis of the health indicator distribution of each group is performed based on the user grouping results, and the optimal warning threshold for each group is determined by ROC curve; The distribution of health indicators can be statistical characteristics of physiological parameters within different user groups, such as mean and standard deviation. These can be defined using statistical methods or domain knowledge. For example, indicators may include resting heart rate and blood oxygen saturation. The ROC curve can be a coordinate graph evaluating the performance of the classification model, with the false positive rate and true positive rate on the horizontal and vertical axes, respectively. For example, by calculating the TPR and FPR at different thresholds, an ROC curve can be plotted, and the threshold corresponding to the maximum AUC or the sensitivity / specificity balance point can be selected. In a specific embodiment, for the elderly hypertension group, the blood pressure value corresponding to an FPR of 5% may be selected as the lower warning limit. This technology reduces false alarms or missed alarms caused by traditional uniform standards by dynamically adjusting the threshold.

[0083] S603: Establish a health knowledge base based on rules and cases, including diagnostic rules in the form of ifthen and typical health cases; The diagnostic rules can be structured decision logic, extracted from expert experience or data mining. For example, a rule could include "If a resting heart rate >90 bpm for three consecutive days and sleep quality is poor, a cardiovascular risk warning is triggered." Typical health cases can be standardized stored user historical data, including fields such as user ID, symptom description, physiological parameters, intervention measures, and results. This technology transforms common patterns in group data into reusable evaluation criteria while retaining the reference value of the cases.

[0084] S604: Construct a case library using knowledge graph technology. The case library includes the relationships between physiological parameters, diseases, symptoms and interventions. Relevant knowledge is retrieved through semantic similarity calculation. The knowledge graph can be a graph-structured network of entity relationships, where nodes represent entities such as diseases and symptoms, and edges represent causal or correlational relationships. For example, the causal relationship between "hypertension" and "high-salt diet" can be extracted through expert annotation or text mining. Semantic similarity calculation can be based on vector embedding or path length quantification methods. For example, cosine similarity can be used to measure the semantic proximity between cases. In a specific embodiment, when retrieving a new case, the system calculates its similarity to nodes in the knowledge graph and returns the most relevant case and its intervention strategy. This technique improves the accuracy of case retrieval by enhancing contextual association analysis capabilities.

[0085] S605: Receives basic information and initial health data from new users, performs rule matching based on the health knowledge base, retrieves similar cases from the case library, and combines Bayesian inference methods to derive personalized health assessment plans and early warning parameters.

[0086] Rule matching can be based on a logical comparison between user characteristics and knowledge base rules. For example, the system filters applicable rules based on information such as age and medical history. Bayesian inference can be a probabilistic update method that combines prior knowledge with real-time data to calculate posterior probabilities. For example, a Bayesian network integrates risk labels from the rule base, probability distributions from the case base, and the user's current data to output the optimal evaluation scheme. In a specific embodiment, if a user's case has a high similarity to a "prediabetes" case and current blood glucose fluctuations are large, the system may increase monitoring frequency and recommend dietary interventions. This technology reduces evaluation bias caused by individual differences or data noise by fusing multi-source data and probabilistic inference.

[0087] This embodiment acquires multi-dimensional user data and uses the K-means algorithm to segment users, dynamically optimizes warning thresholds using ROC curves, constructs a structured knowledge system by combining a rule base and a case base, enhances case association analysis capabilities through knowledge graphs, and finally integrates Bayesian inference to generate personalized assessment solutions. This significantly improves the group adaptability and individual accuracy of health assessments. This method reduces false alarm rate differences through segmentation and threshold optimization, addresses complex health scenarios through the collaboration of a knowledge base and case base, reduces assessment bias through Bayesian inference, and balances standardized solutions with dynamic adaptation requirements, thereby enhancing the intelligence and scalability of the health management system.

[0088] In one embodiment, user health monitoring data, early warning records, and intervention effects are integrated into a health management platform for comprehensive management. This platform tracks changes in user health status in real time, assesses key health indicators and potential risks, and generates health reports based on the monitoring results, including: The system receives health monitoring service requests carrying user identifiers, retrieves user health records and historical monitoring data based on the service requests, and generates personalized monitoring plans based on the health records and historical monitoring data. The health monitoring service request can be a data interaction request with authentication information initiated by the user or medical terminal, such as a JSON request packet submitted through an API interface, containing parameters such as user ID and monitoring type. User health records can be stored in a distributed database and obtained by associating them with user identifiers. For example, a user health record may include personalized medical information such as allergy history and medication records. Historical monitoring data can be retrieved through a time-series database. For example, historical monitoring data may include physiological parameter records such as heart rate variability curves over the past three months. Personalized monitoring solutions can be dynamic combinations of monitoring parameters generated based on a preset rule engine or machine learning model. For example, personalized monitoring solutions may include specific configurations such as adjusting the frequency of blood oxygen monitoring or increasing the frequency of nighttime blood pressure monitoring.

[0089] The technical operation can be achieved through steps such as identity verification, data retrieval, and scheme generation. For example, the system first verifies user permissions through the OAuth2.0 protocol, then retrieves health records and historical monitoring data associated with the user ID from a distributed database, and finally generates personalized monitoring parameters based on the comprehensive data features using a decision tree algorithm. This process can achieve dynamic adaptation of monitoring schemes, such as increasing the frequency of nighttime blood pressure monitoring for hypertensive patients.

[0090] Acquire physiological parameters, exercise data and warning events sent in real time by smart wearable devices, track key health indicators and warning frequency in real time, and determine whether physiological parameters and exercise data are within the normal range and whether warning events are handled in a timely manner. Physiological parameters can be real-time biometric data collected by smart wearable devices, such as resting heart rate and blood oxygen saturation. Exercise data can be activity intensity, step count, and other exercise-related parameters recorded by the device's sensors. Warning events can be abnormal physiological states or risk alerts detected by the device, such as atrial fibrillation warnings or high blood sugar alerts. Key health indicators can be defined through preset individualized threshold ranges; for example, key health indicators can include abnormal states such as a resting heart rate exceeding 100 bpm. Warning frequency can be obtained through time window statistics; for example, warning frequency can record the number of high-risk warnings triggered within 24 hours.

[0091] Technical operation can be achieved through sub-steps such as threshold comparison, timeliness assessment, and status marking. For example, the system continuously receives raw data streams reported by wearable devices, calculates real-time indicators such as the current average resting heart rate through a sliding time window, and compares them with individualized warning thresholds. Simultaneously, it records the timestamp of the warning event trigger. If the user does not confirm or handle a high-risk warning within a preset time (e.g., within 2 hours), it is marked as "unhandled." This process can establish a multi-dimensional health status assessment mechanism, such as monitoring both the degree of deviation in physiological data and assessing the user's response to warnings.

[0092] If physiological parameters exceed the normal range, exercise data is abnormal, or warning events are not handled in a timely manner, health reminders will be automatically sent to the user and authorized guardian through the message push engine. The message push engine can be a rule-based or AI-driven automated notification system, such as using Firebase Cloud Messaging for cross-platform push notifications. Health tips can be text messages containing specific advice; for example, a health tip might include instructions such as "It is recommended to rest immediately and measure your blood pressure." Authorized guardians can be determined through a pre-authorization process; for example, authorized guardians could be emergency contacts or medical caregivers.

[0093] Technical operations can be implemented through conditional triggering, content generation, and multi-channel transmission. For example, when a resting heart rate >100 bpm is detected for three consecutive hours or an unprocessed warning is triggered, the system calls a template engine to generate a text message, which is simultaneously sent to the user and pre-authorized emergency contacts via device vibration, app notifications, SMS, and other channels. This process can build a multi-level response mechanism, ensuring timely delivery of warning information and reducing the possibility of users ignoring risks.

[0094] Receive health report generation instructions, and automatically generate periodic health reports and annual health assessment reports based on the instructions, user health trends, early warning records, and intervention effects; The health report generation command can be a request initiated by the user or medical institution, or a batch processing task automatically triggered by the system based on a preset cycle (e.g., the 1st of each month). User health trends can be obtained through time series analysis; for example, user health trends can include changes in sleep quality scores over the past 30 days. Warning records can be obtained through event log statistics; for example, warning records can include the number of high heart rate warnings this month. Intervention effects can be obtained by comparing data before and after the intervention; for example, intervention effects can include changes in blood pressure after the user followed exercise recommendations.

[0095] Technical operations can be achieved through steps such as data aggregation, trend analysis, and visualization rendering. For example, the system extracts monitoring data, early warning event statistics, and intervention implementation records for a specified time period from the database. Then, it predicts health trends using an ARIMA model and generates visual reports including line charts and heatmaps using Echarts or Tableau. This process can transform scattered monitoring data into structured evaluation conclusions, such as judging the effectiveness of antihypertensive intervention programs by comparing annual blood pressure fluctuation trends.

[0096] This embodiment receives health monitoring service requests carrying user identifiers and generates personalized monitoring plans. It tracks physiological parameters, exercise data, and the status of early warning events in real time. Based on multi-dimensional assessment, it triggers multi-channel health prompts and automatically generates structured health reports by combining health trends, early warning records, and intervention effects. This achieves the technical effects of dynamically adapting monitoring plans, establishing a multi-dimensional risk assessment mechanism, transforming raw data into predictive assessment conclusions, and constructing a hierarchical response system. Ultimately, it enables the health management platform to have intelligent decision-making capabilities, strengthens the individual's proactive management of health risks, provides reliable basis for medical decisions, and promotes the development of precision health management towards intelligence and service.

[0097] Furthermore, this embodiment of the invention also proposes a storage medium storing an AI-based intelligent wearable health monitoring program, which, when executed by a processor, implements the steps of the AI-based intelligent wearable health monitoring method described above.

[0098] In addition, refer to Figure 5 This invention also proposes an artificial intelligence-based intelligent wearable health monitoring system, which includes: The data modeling module 10 is used to acquire the user's multimodal physiological parameters and behavioral data through a smart wearable device, determine the user's health indicators based on the multimodal physiological parameters, and establish a nonlinear mapping relationship between the user's health indicators and the behavioral data to obtain a health status model. Parameter optimization module 20 is used to optimize and solve the personalized health warning threshold of the target user based on the constructed health status model, so as to obtain the target warning parameters suitable for the health status of the target user; The real-time early warning module 30 is used to configure the optimized target early warning parameters to the smart wearable device, collect the target user's physiological parameters in real time and perform dynamic monitoring, and trigger a health risk early warning mechanism when the physiological parameters are abnormal. The trend intervention module 40 is used to perform trend analysis on the physiological parameters and behavioral data of target users during daily monitoring, and to generate health intervention suggestions for abnormal fluctuations or potential health risks. The model correction module 50 is used to acquire users' historical health data and medical diagnosis results, perform big data analysis on physiological parameters, exercise behavior data and historical health data, and continuously correct and improve the health status model based on the inherent relationship between the data. The knowledge construction module 60 is used to obtain health status models under the characteristics of different user groups, establish a health knowledge base and case library for personalized health monitoring, and form a standardized health assessment scheme. The comprehensive management module 70 is used to integrate user health monitoring data, early warning records, and intervention effects into the health management platform for comprehensive management, track changes in user health status in real time, assess key health indicators and potential risks, and generate health reports from the monitoring results.

[0099] Other embodiments or specific implementations of the AI-based intelligent wearable health monitoring system described in this invention can be found in the above-described method embodiments, and will not be repeated here.

[0100] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0101] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments. In the module claims listing several systems, several of these systems may be specifically embodied by the same hardware item. The use of the terms first, second, and third, etc., does not indicate any order and can be interpreted as names.

[0102] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as a read-only memory image (ROM) / random access memory (RAM), magnetic disk, optical disk), and includes several instructions to cause a terminal user device (which may be a mobile phone, computer, server, air conditioner, or network user device, etc.) to execute the methods described in the various embodiments of the present invention.

[0103] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An artificial intelligence-based intelligent wearable health monitoring method, characterized by, The method comprises: obtaining multi-modal physiological parameters and behavior data of a user through a smart wearable device, determining a user health index according to the multi-modal physiological parameters, and establishing a nonlinear mapping relationship between the user health index and the behavior data to obtain a health state model; optimizing and solving a personalized health warning threshold of a target user according to the constructed health state model to obtain a target warning parameter suitable for the health status of the target user; configuring the optimized target warning parameter to the smart wearable device, collecting and dynamically monitoring physiological parameters of the target user in real time, and triggering a health risk warning mechanism when the physiological parameters are abnormal; analyzing the trend of the physiological parameters and behavior data of the target user in the daily monitoring process, and generating health intervention suggestions for abnormal fluctuations or potential health risks; obtaining user historical health data and medical diagnosis results, performing big data analysis on physiological parameters, exercise behavior data and historical health data, and continuously correcting and improving the health state model according to the internal relationship between the data; obtaining health state models under different user group characteristics, establishing a health knowledge base and case base for personalized health monitoring, and forming a standardized health assessment scheme; integrating user health monitoring data, warning records and intervention effects into a health management platform for comprehensive management, tracking the changes of the user's health status in real time, evaluating key health indicators and potential risks, and forming a health report based on the monitoring results. 2.The AI-based smart wear-type health monitoring method of claim 1, wherein The multi-modal physiological parameters include heart rate, blood oxygen saturation and skin temperature; the behavior data includes step count, exercise type, exercise intensity and sleep structure; the method of obtaining multi-modal physiological parameters and behavior data of a user through a smart wearable device, determining a user health index according to the multi-modal physiological parameters, and establishing a nonlinear mapping relationship between the user health index and the behavior data to obtain a health state model comprises: obtaining multi-modal physiological parameters and behavior data of a user through a smart wearable device, and preprocessing the multi-modal physiological parameters and behavior data, including data cleaning, normalization and feature extraction; determining a user health index according to the preprocessed multi-modal physiological parameters, including a cardiovascular health score, a metabolic health score and a sleep health score; establishing a nonlinear mapping relationship between the user health index and the behavior data through a long short-term memory network algorithm to obtain a health state model. 3.The AI-based smart wear-type health monitoring method of claim 1, wherein The health warning threshold includes a resting heart rate abnormal threshold, an active heart rate upper limit and a blood oxygen saturation lower limit; the genetic algorithm intelligent optimization algorithm is used to optimize the health warning threshold of the target user according to the constructed health state model, and the optimization includes the following steps: According to a preset chromosome coding mode, the resting heart rate abnormal threshold, the active heart rate upper limit and the blood oxygen saturation lower limit are coded into binary strings, and the length of the binary string is set according to the parameter accuracy requirement; obtaining the set binary string, setting the fitness function of the genetic algorithm, and converting the behavior data into dimensionless indicators by using a normalization processing method; After the fitness function is set, an excellent individual is selected by using a roulette wheel selection operation, a crossover probability of the roulette wheel selection operation is set as a preset value, and gene recombination is performed by using a single-point crossover mode; According to the selected excellent individual, the genetic algorithm is iteratively run, a population size of the genetic algorithm is set as a first preset value, and a maximum iteration number is set as a second preset value; When the genetic algorithm reaches a convergence condition or the maximum iteration number, a target user individualized health early warning threshold is obtained by decoding, and is used as a target early warning parameter applicable to a health condition of the target user; The fitness function formula of the target user individualized health early warning threshold is: ； In the formula, is the fitness function, is the resting heart rate abnormal threshold, is the upper limit of the activity heart rate, is the lower limit of blood oxygen saturation, is the weight of the kth health indicator, satisfying , is the health state model output value based on the current threshold , is the actual measured health indicator value, is the theoretical maximum value of the kth health indicator, is the theoretical minimum value of the kth health indicator. 4.The AI-based smart wear-type health monitoring method of claim 3, wherein After the genetic algorithm reaches the convergence condition or the maximum iteration number, the target user individualized health early warning threshold is obtained by decoding, and is used as the target early warning parameter applicable to the health condition of the target user, the method further includes: The multiple modal physiological parameters are encoded as an individual by using a parameter combination, and an initial population is randomly generated; The fitness of each individual is calculated based on a health state model, an excellent individual is selected, and a new offspring individual is generated by using crossover and mutation; The parameter combination is optimized by iteratively performing evolution operations multiple times, and the individual with the highest fitness is finally selected, and the parameter combination corresponding to the individual is used as the target early warning parameter. 5.The AI-based smart wear-type health monitoring method of claim 1, wherein The target early warning parameter obtained by optimization is configured to the smart wearable device, physiological parameters of the target user are collected in real time and dynamically monitored, and a health risk early warning mechanism is triggered when the physiological parameters are abnormal, including: The target early warning parameter obtained by optimization is acquired, and the target early warning parameter is downloaded to a local processor of the smart wearable device through a Bluetooth low energy protocol; The heart rate and blood oxygen saturation data of the user are continuously collected by using a multi-sensor fusion technology, and the optimal estimated physiological parameters of the target user are obtained by using a particle filtering algorithm; The change rate of the optimal estimated physiological parameters of the target user is calculated by using a sliding window method, and abnormal detection is performed in combination with the target early warning parameter; If the heart rate of the target user is sampled for three times continuously and is out of the target interval, or the blood oxygen saturation is lower than a lower limit of the early warning, the health risk early warning mechanism is triggered, and warning information is pushed to the user terminal. 6.The AI-based smart wear-type health monitoring method of claim 1, wherein In the daily monitoring process, trend analysis is performed on the physiological parameters and behavior data of the target user, and a health intervention suggestion is generated for abnormal fluctuations or potential health risks, including: In the daily monitoring process, a time series decomposition method is used to monitor the long-term change trend of the physiological parameters of the target user, the time series decomposition method applies a trend term, a periodic term, and a random term separation model; It is judged whether the physiological parameters of the target user have an abnormal trend according to the trend term, the periodic term, and the random term separation model; If there is no abnormal trend, it is determined that the health state of the user is stable, and the routine monitoring is continued; If there is an abnormal trend, it is determined that the health state of the user has a potential risk, a health intervention suggestion generation program is triggered, and a health intervention suggestion is pushed in combination with the user's exercise behavior data, the health intervention suggestion includes exercise intensity adjustment, work-rest optimization, and diet recommendation. 7.The AI-based smart wear-type health monitoring method of claim 1, wherein The user historical health data includes previous medical history, medication records, and physical examination reports. The user historical health data includes previous medical history, medication records, and physical examination reports. The user historical health data includes previous medical history, medication records, and physical examination reports. The user historical health data includes previous medical history, medication records, and physical examination reports. According to the fused data, a feature engineering method is used for dimension expansion to extract time domain features, frequency domain features, and nonlinear features. Using the extracted time domain features, frequency domain features, and nonlinear features, a transfer learning algorithm is used to update the health status model, and the association rules between the physiological parameters, the motion behavior data, and the disease occurrence risk are mined to obtain the rule "if the nocturnal heart rate variability decreases and the deep sleep proportion decreases, then the cardiovascular risk increases". An online learning method is used to correct the health status model, and an incremental gradient descent method is used to update the health status model parameters to adapt to the dynamic changes of the user's health status. 8.The AI-based smart wear-type health monitoring method of claim 1, wherein The health status model under different user group characteristics is obtained, the health knowledge base and case base of individualized health monitoring are established, and the standardized health assessment scheme is formed, including: According to the health status model under different user group characteristics, the monitoring data and model evaluation results of users of different ages, genders, and health conditions are obtained, and a K-means algorithm is used for user grouping. According to the user grouping results, the health index distribution of each group is statistically analyzed, and the optimal early warning threshold of each group is determined through the ROC curve. A health knowledge base based on rules and cases is established, including diagnostic rules in the form of if-then and typical health cases. A case base is constructed using knowledge graph technology, including the association between physiological parameters, diseases, symptoms, and intervention measures, and relevant knowledge is retrieved through semantic similarity calculation. The basic information and initial health data of new users are received, rule matching is performed according to the health knowledge base, similar cases are retrieved from the case base, and individualized health assessment scheme and early warning parameters are obtained by combining Bayesian inference method. 9.The AI-based smart wear-type health monitoring method of claim 1, wherein The user health monitoring data, early warning records, and intervention effects are integrated into the health management platform for comprehensive management, real-time tracking of user health status changes, assessment of key health indicators and potential risks, and formation of health reports from monitoring results, including: The health monitoring service request carrying the user identifier is received, the user health record and historical monitoring data are retrieved according to the service request, and the health management platform generates an individualized monitoring scheme based on the health record and the historical monitoring data. The physiological parameters, motion data and early warning events sent by the smart wearable device in real time are acquired, and key health indicators and early warning frequencies are tracked in real time to determine whether the physiological parameters, motion data are within a normal range and early warning events are handled in time; If the physiological parameters are outside the normal range, the motion data is abnormal or early warning events are not handled in time, a health prompt is automatically sent to the user and authorized guardians through a message push engine; A health report generation instruction is received, and a periodic health report and an annual health assessment report are automatically generated according to the generation instruction, user health trends, early warning records and intervention effects.

10. An artificial intelligence based smart wearable health monitoring system characterized in that, The intelligent wearable health monitoring system based on artificial intelligence comprises: A data modeling module is configured to acquire multi-modal physiological parameters and behavior data of a user through a smart wearable device, determine user health indicators according to the multi-modal physiological parameters, and establish a nonlinear mapping relationship between the user health indicators and the behavior data to obtain a health state model; A parameter optimization module is configured to optimize and solve a health early warning threshold of a target user according to the constructed health state model to obtain a target early warning parameter suitable for the health status of the target user; A real-time early warning module is configured to configure the optimized target early warning parameter to the smart wearable device, collect physiological parameters of the target user in real time and perform dynamic monitoring, and trigger a health risk early warning mechanism when the physiological parameters are abnormal; A trend intervention module is configured to perform trend analysis on physiological parameters and behavior data of the target user during daily monitoring, and generate health intervention suggestions for abnormal fluctuations or potential health risks; A model correction module is configured to acquire historical health data and medical diagnosis results of the user, perform big data analysis on physiological parameters, motion behavior data and historical health data, and continuously correct and improve the health state model according to the internal relationship between the data; A knowledge construction module is configured to acquire health state models under different user group characteristics, establish a health knowledge base and a case base for personalized health monitoring, and form a standardized health assessment scheme; A comprehensive management module is configured to integrate user health monitoring data, early warning records and intervention effects into a health management platform for comprehensive management, track changes in user health status in real time, evaluate key health indicators and potential risks, and form health reports from monitoring results.

Images