Health monitoring bracelet based on multi-modal data fusion and knowledge graph

Abstract

The invention relates to the technical field, in particular to a health monitoring bracelet based on multi-modal data fusion and a knowledge graph. Comprising a physiological data acquisition module, a feature extraction module, a knowledge graph construction module, a trend modeling module and a rule judgment module. According to the invention, through time alignment and fluctuation segment calibration of multi-source physiological data, in combination with synchronization and coordination analysis between features, association expression ability of multi-dimensional physiological features is enhanced, health standards and index combinational logic are adopted to construct health portraits, and identification precision of individual states is optimized. Modeling is carried out by using a combined trend of an activity state and a heart load, dynamic perception of a health state change is realized, a dynamic threshold adjustment mechanism is used, sensitivity of abnormal event detection is improved, a health risk level is assessed in combination with abnormity continuity and trend characteristics, accuracy and an intervention reference value of a risk assessment result are enhanced, and the risk assessment efficiency is improved. And the intelligent level of health management is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent health management technology, and in particular to a health monitoring wristband based on multimodal data fusion and knowledge graph. Background Technology

[0002] The field of intelligent health management technology encompasses the entire process of health information collection, data fusion, intelligent analysis, and health intervention for individuals or specific populations. Its core lies in the real-time collection of physiological parameters and behavioral data using various hardware devices such as wearable devices, biosensors, and IoT terminals. This is combined with edge computing and cloud computing architectures to conduct multi-dimensional analysis and dynamic modeling of health data. Furthermore, through artificial intelligence algorithms and knowledge graphs, individual health profiles are constructed to achieve health status monitoring, disease risk assessment, and intervention strategy recommendations. The field of intelligent health management technology covers multimodal data fusion processing, health trend prediction, risk grading assessment, personalized intervention strategy formulation, and secure storage and privacy protection of health information. It achieves full-process technology integration and collaboration from data perception, intelligent processing, knowledge management to service execution.

[0003] Among them, the health monitoring bracelet based on multimodal data fusion and knowledge graph refers to a wearable device system that integrates multiple physiological and behavioral signal acquisition methods, combined with knowledge graph construction and rule reasoning mechanisms, to achieve continuous perception and dynamic management of the user's health status. The patent topic mainly involves key technical aspects such as multimodal acquisition and fusion processing of physiological signals, identification and analysis of abnormal states, construction and reasoning of health knowledge graphs, and secure transmission and storage of data. Specifically, it includes using Kalman filtering for signal fusion, wavelet transform for signal reconstruction, edge AI models to perform local initial screening, Rete rule engine for decision reasoning, and blockchain and encryption protocols to ensure data compliance and privacy security.

[0004] Traditional health monitoring bracelets rely on single physiological parameters or a small number of feature values ​​for actual health monitoring. They lack dynamic calibration and stability enhancement for multi-source heterogeneous data, making them prone to distortion or loss when data quality is compromised. This results in insufficient continuity and authenticity of monitoring results. In feature recognition, they judge changes based on single indicators, failing to effectively characterize the correlation between multiple indicators. They lack a multi-dimensional and dynamic expression of individual health status, resulting in health profiles that are general but lack personalization. In health status modeling, they rely heavily on static rules or fixed thresholds for anomaly detection, lacking analysis of status change trends and dynamic threshold adjustment. This leads to misjudgments or missed detections when environmental changes or physiological fluctuations occur. In health risk assessment, they assess levels based on the number of abnormal events or the severity of a single event, failing to combine the duration, frequency, and health trend changes of abnormal states for multi-dimensional comprehensive analysis. This results in insufficient adaptability and scientific rigor of health risk assessment results, affecting the rationality of intervention strategies. Summary of the Invention

[0005] To address the technical problems existing in the prior art, this invention provides a health monitoring wristband based on multimodal data fusion and knowledge graph. The technical solution is as follows: On the one hand, a health monitoring wristband based on multimodal data fusion and knowledge graph is provided. The system includes: The physiological data acquisition module calls the wristband sensor to collect various physiological and acceleration data. Through timestamps, it performs time alignment and data interpolation on the data, identifies and calibrates abnormal fluctuation data segments, and generates a multi-source fusion dataset. Based on the multi-source fusion dataset, the feature extraction module extracts the fluctuation amplitude, fluctuation frequency, and stability duration of each physiological parameter, analyzes the synchronicity of heart rate and blood oxygen, assesses the coordination of skin conductance and temperature, and generates comprehensive feature index values. Based on the comprehensive feature index values ​​and combined with preset health standards, the knowledge graph construction module identifies health level ranges for multiple feature values. By analyzing and identifying the combination logic between multiple indicators, it constructs an individualized structured health profile and generates a health knowledge graph dataset. The trend modeling module, based on the health knowledge graph dataset, calls continuous time series of acceleration, skin conductance, and heart rate data to analyze the user's activity intensity and detect fall events. It combines the fluctuation frequency of heart rate data in various activity states to assess the user's cardiac load status, extracts the user's health change characteristics over continuous time periods, and obtains health status trend values.

[0006] As a further aspect of the present invention, the multi-source fusion dataset includes the original sensor dataset, time-aligned records, and dataset interpolation processing results. The comprehensive feature index values ​​include the amplitude values ​​of physiological parameter change intervals, the synchronization degree of fluctuations between parameters, the stable duration of resting state, and the degree of skin signal coupling. The health knowledge graph dataset specifically includes individual physiological parameter health level identifiers, cross-parameter index combination patterns, and structured health status nodes. The health status trend values ​​include the amplitude of activity intensity changes, the fall event recognition status, and the continuous change trajectory of health characteristics.

[0007] As a further aspect of the present invention, the physiological data acquisition module includes: The data acquisition submodule calls the wristband sensor to acquire the user's heart rate data, blood oxygen data, skin temperature data, skin conductance data, and real-time acceleration data, and establishes the raw sensor dataset; The data alignment submodule, based on the original sensor dataset, calls the timestamp of each data item to perform time alignment on the data, and combines data interpolation to supplement the data for missing time periods, thereby establishing a time-aligned dataset. The data calibration submodule, based on the time-aligned dataset, calls the change trend of each data item, calculates the fluctuation amplitude of heart rate data, blood oxygen data, skin temperature data, and skin conductance data, identifies abnormal fluctuation data segments and performs calibration, and obtains a multi-source fusion dataset.

[0008] As a further aspect of the present invention, the feature extraction module includes: The data feature calculation submodule extracts the numerical variation range of each physiological parameter based on the multi-source fusion dataset, calculates the amplitude of the numerical change of each data in a continuous period, performs frequency statistics on the number of changes of data within a time window, obtains the stable duration of the numerical change of each data, and establishes basic feature parameter values. The feature association submodule calls the basic feature parameter values ​​to perform synchronization calculations on the changing trends of heart rate data and blood oxygen data, calculates the time difference of data change peaks, and obtains the physiological synchronization correlation degree. The feature statistics submodule evaluates the coordination of the changing trends of skin electrodermatology data and skin temperature data based on the physiological synchronization correlation, calculates the difference rate of data change magnitude, and establishes a comprehensive feature index value.

[0009] As a further aspect of the present invention, the knowledge graph construction module includes: The health interval identification submodule identifies health level intervals for multiple feature values ​​by comparing them with preset health standards based on the comprehensive feature index values, and establishes level interval coefficients. The indicator combination analysis submodule calls the level interval coefficient, compares the fluctuation direction of multiple feature values ​​within the same time period, analyzes the consistency of waveform trends, identifies the combination logic between multiple indicators, and obtains the indicator logic matching rate. The health profile construction submodule calls the logical matching rate of the indicators, analyzes the distribution of the level range of each feature value, evaluates the correlation level between multiple sets of indicators and defines the individualized node structure, constructs an individualized structured health profile, and establishes a health knowledge graph dataset.

[0010] As a further aspect of the present invention, the trend modeling module includes: The behavior recognition submodule, based on the health knowledge graph dataset, calls the continuous time series of acceleration data and skin conductance activity data to calculate the user's activity intensity in real time, detect fall events, and generate activity status values. The load assessment submodule calls the activity status value, combines it with the time series of continuous heart rate data, extracts the heart rate variation amplitude and frequency corresponding to various activity states, assesses the user's cardiac load status in real time, and obtains the cardiac load distribution rate. The trend extraction submodule, based on the cardiac load distribution rate, compares the user's activity intensity and cardiac load to assess and analyze the health change characteristics over a continuous period of time in real time, and establishes a health status trend value.

[0011] As a further aspect of the present invention, the specific formula for calculating the user's activity intensity in real time is as follows: ; Calculate the user's activity intensity value; in, For the first The user's activity intensity value at any given time. For the first The acceleration value in the X-axis direction at time 1. For the first The acceleration value in the Y-axis direction at time 1. For the first The acceleration value in the Z-axis direction at time 1. For the first skin electrical activity conductivity value at any time For the first skin electrical activity conductivity value at any time For the first The sampling time point at time . For the first The sampling time point before the previous time. The normalized coefficients for the acceleration component are... This is the normalization coefficient for the change in skin electrical activity.

[0012] As a further aspect of the present invention, the system further includes: Based on the health status trend value, the rule judgment module uses an individualized structured health profile to adjust the blood oxygen threshold and respiratory rate abnormality detection threshold, detects abnormal blood oxygen data and respiratory rate data in real time, extracts the time period of abnormal status, combines the user's health change characteristics, calculates the user's health risk level, and generates a health risk assessment value. The health risk assessment values ​​specifically refer to blood oxygen risk level, abnormal respiratory rate interval label, and abnormal distribution coefficient of health trend.

[0013] As a further aspect of the present invention, the rule judgment module includes: The threshold adjustment submodule, based on the health status trend value, calls the distribution range of individual physiological parameters in the structured health profile, calculates the distribution offset of blood oxygen value and respiratory rate value in multiple level intervals under the current user status, adjusts the blood oxygen detection threshold and respiratory rate abnormal judgment threshold according to the offset direction, and generates an individualized monitoring threshold group. The anomaly extraction submodule calls the individualized monitoring threshold group to compare real-time blood oxygen data and respiratory rate data, detects abnormal data and locates the time period of the abnormal state, and obtains the abnormal state interval value. The risk assessment submodule calculates the user's health risk level in real time based on the abnormal state range value and the characteristics of health changes, and establishes a health risk assessment value.

[0014] As a further aspect of the present invention, the specific formula for calculating the distribution offset of the user's current blood oxygen value and respiratory rate value within multiple level intervals is as follows: ; Calculate the offset equalization value; in, This represents the joint offset equilibrium value of the distribution of blood oxygen and respiratory rate within the current user's healthy range. This represents the average blood oxygen level during the current monitoring period. This represents the median value of an individual's blood oxygen saturation range recorded in the health profile. Represents the upper limit of an individual's blood oxygenation range. Represents the lower limit of an individual's blood oxygenation range. This represents the average respiratory rate during the current monitoring period. This represents the median value of the individual's respiratory rate range recorded in the health profile. Represents the upper limit of an individual's respiratory rate range. This represents the lower limit of an individual's respiratory rate range.

[0015] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: By aligning and calibrating the fluctuation segments of multi-source physiological data over time, and combining the analysis of synchronicity and coordination among features, the correlation expression ability of multidimensional physiological features is enhanced. A health profile is constructed using a combination logic of health standards and indicators, optimizing the identification accuracy of individual status. By using joint trend modeling of activity status and cardiac load, dynamic perception of changes in health status is achieved. A dynamic threshold adjustment mechanism is used to improve the sensitivity of abnormal event detection. By combining the persistence of abnormalities and trend characteristics to assess health risk levels, the accuracy of risk assessment results and the reference value for intervention are enhanced, thereby improving the level of intelligence in health management. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a system flowchart of the present invention; Figure 2 This is a schematic diagram of the system framework of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0019] In embodiments of the present invention, words such as "exemplarily," "for example," etc., are used to indicate that something is an example, illustration, or description. Any embodiment or design described as "exemplary" in the present invention should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of the word "exemplary" is intended to present the concept in a concrete manner. Furthermore, in embodiments of the present invention, the meaning expressed by "and / or" can be both, or either one.

[0020] In the embodiments of this invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent. Similarly, the terms "of," "corresponding (relevant)," and "corresponding" may sometimes be used interchangeably. It should be noted that, without emphasizing the distinction between them, their intended meanings are consistent.

[0021] In this embodiment of the invention, sometimes a subscript such as W1 may be written in a non-subscript form such as W1. When the difference is not emphasized, the meaning they express is the same.

[0022] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0023] This invention provides a health monitoring wristband based on multimodal data fusion and knowledge graph. Please refer to [link / reference]. Figures 1 to 2 This invention discloses a health monitoring wristband based on multimodal data fusion and knowledge graph, comprising: The physiological data acquisition module calls the wristband sensor to collect various physiological and acceleration data. Through timestamps, it performs time alignment and data interpolation on the data, identifies and calibrates abnormal fluctuation data segments, and generates a multi-source fusion dataset. The feature extraction module is based on a multi-source fusion dataset to extract the fluctuation amplitude, fluctuation frequency, and stability duration of each physiological parameter, analyze the synchronicity of heart rate and blood oxygen, assess the coordination of skin conductance and temperature, and generate comprehensive feature index values. The knowledge graph construction module identifies health level ranges for multiple feature values ​​based on comprehensive feature index values ​​and pre-set health standards. By analyzing and identifying the combination logic between multiple indicators, it constructs an individualized structured health profile and generates a health knowledge graph dataset. The trend modeling module is based on a health knowledge graph dataset. It calls continuous time series of acceleration, skin conductance, and heart rate data to analyze the user's activity intensity and detect fall events. It combines the fluctuation frequency of heart rate data in various activity states to assess the user's cardiac load status, extract the user's health change characteristics in continuous time periods, and obtain health status trend values. The rule-based judgment module uses health status trend values ​​and individualized structured health profiles to adjust blood oxygen threshold and respiratory rate abnormality detection thresholds. It detects abnormal blood oxygen data and respiratory rate data in real time, extracts the time period of abnormal status, and calculates the user's health risk level by combining the user's health change characteristics, generating a health risk assessment value.

[0024] The multi-source fusion dataset includes the original sensor dataset, time-aligned records, and dataset interpolation results. The comprehensive feature index values ​​include the amplitude of physiological parameter changes, the synchronization of fluctuations between parameters, the duration of stable resting state, and the degree of skin signal coupling. The health knowledge graph dataset specifically includes individual physiological parameter health level identifiers, cross-parameter indicator combination patterns, and structured health status nodes. The health status trend values ​​include the amplitude of activity intensity changes, fall event recognition status, and continuous change trajectory of health characteristics. The health risk assessment values ​​specifically refer to the blood oxygen risk level, abnormal respiratory rate interval labels, and abnormal distribution coefficient of health trends.

[0025] The physiological data acquisition module includes: The data acquisition submodule calls the wristband sensor to acquire the user's heart rate data, blood oxygen data, skin temperature data, skin conductance data, and real-time acceleration data, and establishes the raw sensor dataset; The data acquisition submodule calls the wristband's sensors to acquire the user's heart rate, blood oxygen, skin temperature, electrodermal skin (EDS) data, and real-time acceleration data. In practical applications, the heart rate value collected every second can be used as a single sample point, and continuous collection can form a one-minute heart rate time series. In this time series, the timestamp, unit time change, and measurement accuracy of each data point are extracted. Skin temperature data is sampled in 0.1℃ increments and recorded every two seconds within the wristband device. EDS data is collected based on the unit change in conductivity, with a corresponding electrode sampling frequency of 5Hz. Real-time acceleration data... The data needs to be broken down into components in the x, y, and z directions, and the magnitude and slope of acceleration in each direction should be recorded. The sampling frequency is set to 10Hz, and the device ID and user ID are associated by number. The unified sampling period is 60000ms. The data collected by various sensors are aggregated into a data block according to the time label and assigned a unique timestamp number for subsequent synchronization. Each data entry in the original sensor dataset contains five types of sensing indicators. Each type of data contains three fields: original measurement value, time label, and device number. For example, the acceleration x-axis data collected at the 15th second is 1.02 m / s², corresponding to a heart rate of 76 bpm, skin temperature of 36.3℃, blood oxygen saturation of 97%, and skin conductivity change of 0.85 μS / cm. The above structure is combined to form a standard structure field, and finally aggregated into a unified sensor data record entry with a timestamp and five types of original data fields to establish the original sensor dataset.

[0026] The data alignment submodule is based on the original sensor dataset. It calls the timestamp of each data item to perform time alignment on the data. Combined with data interpolation, it fills in the data for missing time periods and establishes a time-aligned dataset. The data alignment submodule, based on the original sensor dataset, calls the timestamp of each data item to perform time alignment. First, it extracts the timestamp sequence from all data fields, obtains the start and end times for each type of sensor data, and calculates the time interval between adjacent samples, using seconds to determine if there are gaps in continuous data. If the interval between two timestamps is greater than the standard value of the sampling period, it is identified as a missing segment. The standard value of the sampling period is set according to the sensor type: heart rate 5s, acceleration 0.1s, skin temperature 2s, skin conductance 0.2s, and blood oxygen 3s. The missing segment is filled using linear interpolation between two known points. For example, on the acceleration x-axis, if t1=1.0s is 1.1 m / s² and t2=1.2s is 1.3 m / s², then the interpolated value at t1=1.1s is (1.1+1.3) / 2=1.2. After interpolating all missing data segments in this manner using m / s², all fields are re-sorted according to a unified baseline timestamp sequence to form a data record set with consistent structure, thus completing the time alignment operation. During the time alignment process, if there is a decimal precision deviation that exceeds the upper limit of the perception range error, such as a skin temperature variation exceeding ±0.3℃, the data segment is marked as unreliable and its data mark value is recorded as 0; otherwise, it is marked as 1. Finally, the time unification and missing data repair of all data fields are completed, and a time-aligned dataset is established.

[0027] The data calibration submodule aligns the dataset according to time, calls the change trend of each data item, calculates the fluctuation amplitude of heart rate data, blood oxygen data, skin temperature data, and skin conductance data, identifies abnormal fluctuation data segments and performs calibration, and obtains a multi-source fusion dataset. The data calibration submodule aligns the dataset with time, retrieves the trend of each data item, calculates the fluctuation amplitude of heart rate, blood oxygen, skin temperature, and skin conductance data, identifies and calibrates abnormal fluctuation segments. In practice, continuous time series values ​​for each type of data need to be extracted first, and a baseline range for fluctuation amplitude needs to be set: heart rate ±15 bpm, blood oxygen ±5%, skin temperature ±0.4℃, and skin conductance ±1.0 μS / cm. Within each time window, the difference between the maximum and minimum values ​​is calculated as the fluctuation amplitude. For example, in the heart rate sequence {76, 79, 83, 105, 78}, the fluctuation amplitude is 105-76=29 bpm, which exceeds the baseline range for heart rate fluctuation, and is therefore marked as an abnormal segment. After identifying the abnormal segment, it is repaired based on its boundary value with the adjacent normal segment, using the boundary average value for replacement. For example, if the starting value of the abnormal segment is 105, the preceding normal segment is 83, and the following normal segment is 78, then 105 is replaced with (83+78) / 2=80.5. If the conductivity jump rate in the bpm data exceeds 2 μS / cm / s, it is also considered a transient anomaly. The median of two adjacent stable segments is used to fill the gaps. After replacing all the abnormal segments, the data integrity and continuity are re-verified, and the trusted state flag field of each data is updated to 1 to obtain a multi-source fusion dataset.

[0028] The feature extraction module includes: The data feature calculation submodule is based on a multi-source fusion dataset. It extracts the range of numerical changes for each physiological parameter, calculates the magnitude of the numerical changes of each data point in a continuous period, performs frequency statistics on the number of changes of data within a time window, obtains the stable duration of the numerical changes of each data point, and establishes basic feature parameter values. The data feature calculation submodule, based on a multi-source fusion dataset, extracts the numerical variation range of each physiological parameter. It calculates the amplitude of the numerical change for each data point within a continuous time period, performs frequency statistics on the number of changes within a time window, and obtains the stable duration of the numerical change for each data point. The extraction of the data variation range is based on four types of physiological data: heart rate, blood oxygen, skin temperature, and skin conductance. Each data point is divided into 30-second segments. Within each segment, the maximum and minimum values ​​are extracted, and their difference is calculated as the variation range value for that segment. For example, if the maximum blood oxygen value in a segment is 98% and the minimum is 93%, the variation range is 5%. The numerical variation amplitude is calculated by accumulating the absolute differences between adjacent data points within each segment and dividing by the total number of data points in that segment to obtain the average variation amplitude. For example, if the heart rate data over 5 seconds are 72, 76, and 78... With heart rates of 75 and 74 beats per minute, the adjacent differences are 4, 2, 3, and 1, respectively. The sum of these differences is 10 divided by 4, yielding an average amplitude of 2.5 beats per minute. Frequency statistics are calculated by accumulating the number of changes exceeding the historical stable range threshold within the same time period. The threshold is set based on the standard deviation. Assuming the historical standard deviation of heart rate at rest is 3, this is used as the basis for judging fluctuation frequency. If the data changes more than 3 times consecutively within a 5-second window, it is counted as frequency 1. The calculation method for the stable duration of numerical changes is the cumulative time length during which the amplitude of continuous data changes is lower than the stable amplitude threshold. The stable amplitude threshold is set to 1 beat per minute, meaning that a difference of less than 1 between adjacent heart rates is considered stable. If this state lasts for 10 seconds, the stable duration is 10 seconds. Finally, the above processing is uniformly output across various physiological data dimensions to form structured field records, establishing basic feature parameter values.

[0029] The feature association submodule calls the basic feature parameter values ​​to perform synchronization calculations on the changing trends of heart rate data and blood oxygen data, calculates the time difference of data change peaks, and obtains the physiological synchronization correlation degree. The feature association submodule calls the basic feature parameter values ​​to perform synchronization calculations on the changing trends of heart rate and blood oxygen data. It calculates the time difference between the peak values ​​of the data changes and obtains the synchronization relationship based on extracting the local extreme points of each parameter. First, it identifies the peak points in the heart rate and blood oxygen data within a 5-second window, marking all local maximum values ​​within this window as peak points. For example, if the heart rate sequence is 72, 74, 76, 75, 73, with a peak of 76, and the blood oxygen sequence is 96, 97, 99, 98, 96, with a peak of 99, then the peak time points for both are recorded as follows: and Then, the difference between the two time points is calculated. If the time difference is within a set synchronization threshold, it is marked as a synchronization event. The default synchronization threshold is 500 milliseconds, or 0.5 seconds. It is 12:00:00.200. A time difference of 400 milliseconds (12:00:00.600) is considered a synchronization event. This process is repeated for each valid peak value within a 10-minute period. The frequency of synchronization events is then calculated, and the synchronization correlation is determined using the following formula: ,in To synchronize the degree of correlation, To synchronize the number of events, For the number of peak pairs that can be compared, for example, if 30 sets of data meet the comparison criteria within 10 minutes, and 18 of them are found to be synchronized, then... To obtain the physiological synchronization correlation.

[0030] The feature statistics submodule evaluates the coordination of the changing trends of skin electrodermatology data and skin temperature data based on the physiological synchronization correlation, calculates the difference rate of data change amplitude, and establishes comprehensive feature index values. The feature correlation submodule first calculates the synchronicity of heart rate and blood oxygenation data by analyzing their trends, extracting their peak points, calculating the time difference between peak occurrences, and statistically analyzing the frequency of synchronization events to obtain the physiological synchronicity correlation (R). This indicator reflects the physiological linkage between heart rate and blood oxygenation. Subsequently, the feature statistics submodule, based on the physiological synchronicity correlation, assesses the coordination of skin conductance and skin temperature data. By calculating the difference rate of change amplitude between the two sets of data within the same time window, it analyzes the degree of coordination in their trends. Finally, the comprehensive feature index integrates the physiological synchronicity correlation and skin conductance-temperature coordination, making the synchronicity analysis results an important reference or weighting basis for coordination assessment, reflecting the linkage relationship of various physiological signals in the overall state, and thus reflecting the individual's physiological state.

[0031] The feature statistics submodule assesses the consistency of changes in the trends of electrodermatology (ED) and skin temperature (ST) data based on physiological synchronicity, and calculates the difference rate of data variation. The process begins by using the fluctuation range of EDT and ST data within the same time window as the basis for analysis. The two sets of data are paired along the same time axis, and the maximum and minimum values ​​of the two data points are extracted for every 5-second interval, with their variation amplitude calculated. For example, if the maximum EDT value is 0.52 microsiemens and the minimum is 0.48, with an amplitude of 0.04, and the maximum ST value is 36.9 degrees Celsius and the minimum is 36.5 degrees Celsius, with an amplitude of 0.4 degrees Celsius, the formula for calculating the difference rate is: ,in For the difference rate, For skin electrical amplitude, Substituting the value into the skin temperature range, we get... A smaller difference rate indicates stronger coordination. The difference rates over multiple time periods are averaged to obtain the coordination strength of skin conductance and skin temperature trends over a period. An average difference rate less than 0.2 is considered high coordination, 0.2 to 0.5 is medium coordination, and above 0.5 is weak coordination. This calculation process ultimately outputs the coordination level and its value as features, establishing a comprehensive feature index value. The comprehensive feature index value includes multi-dimensional raw and calculated features, such as heart rate fluctuation amplitude, blood oxygen fluctuation frequency, skin temperature and skin conductance fluctuations and coordination, and physiological synchronicity. Each indicator is compared with preset health standards to classify low / medium / high risk levels, and uniformly coded and summarized to form a level interval coefficient matrix for health interval identification, thus providing a quantitative basis for comprehensive assessment and dynamic early warning of individual health status.

[0032] The knowledge graph construction module includes: The health interval identification submodule identifies health level intervals based on comprehensive feature index values ​​by comparing them with preset health standards, and establishes level interval coefficients. The application of comprehensive feature index values ​​in the health interval identification submodule is reflected in the fact that it is included as a high-level feature in the comparison and judgment of health risk levels, and is input together with individual indicators (such as heart rate fluctuation amplitude, blood oxygen frequency, skin temperature stability duration, etc.) and compared with preset health standards. It can also be used as a weight or correction factor to adjust the risk judgment of individual indicators. For example, when the comprehensive coordination is high, the risk level of some individual indicators can be appropriately lowered, and vice versa. This enables a more scientific and comprehensive classification and judgment of the overall health status of an individual's multidimensional physiological signals, so that the division of health intervals not only depends on the interval judgment of a single parameter, but also reflects the physiological linkage and true risk level after the fusion of multiple indicators.

[0033] The health zone identification submodule identifies health level zones for various characteristic values ​​by comparing them with preset health standards based on comprehensive characteristic index values. This process first takes characteristic index values ​​as core data input, including heart rate fluctuation amplitude, blood oxygen frequency, and skin temperature stability duration. For each characteristic value, a health standard zone is set: low risk zone is 0-5 times per minute, medium risk zone is 5-10 times per minute, and high risk zone is more than 10 times per minute.

[0034] If a user's heart rate fluctuates at 8.3 beats per minute, it falls within the medium-risk range. Similarly, blood oxygen fluctuation frequency is divided into: low risk 0-1 times per minute, medium risk 2-3 times per minute, and high risk greater than 3 times per minute. If the monitoring frequency is 2 times per minute, it is considered medium risk. The duration of stable skin temperature is set with a 5-minute window. If it remains stable for more than 3 minutes, it is considered low risk; if it is less than 2 minutes, it is considered high risk. Data comparison is conducted item by item, comparing the user's current feature value with the upper and lower limits of the standard range. The range boundary is represented as a closed range, and a tolerance value of no more than 1% is set in the boundary judgment. Data within the tolerance range is included in the adjacent lower risk level. After the level judgment of all feature values ​​is completed, different risk levels are uniformly encoded as numerical levels, such as low risk as 1, medium risk as 2, and high risk as 3. The results are stored in the corresponding feature fields and a correlation matrix is ​​generated. The distribution of the current comprehensive index in multiple ranges is transformed through the matrix column vectors and packaged for output, establishing the level range coefficient. An association matrix is ​​a structured data table used to integrate and encode the risk levels of multiple health characteristics. Specifically, it stores the risk level (usually 1 / 2 / 3, representing low / medium / high risk, respectively) of each input feature (such as heart rate fluctuation amplitude, blood oxygen fluctuation frequency, skin temperature stability duration, skin conductance amplitude, and overall coordination) within the current analysis period in matrix form. Each row represents a feature, and each column represents the risk level assessment result for a monitoring period or analysis window. By extracting the column vectors from the matrix, the distribution of all features across different risk intervals within a given time period can be obtained. For example, a column vector of [2, 1, 2, 3, 2] for a 5-second window indicates that heart rate is at medium risk, blood oxygen is low risk, skin temperature is medium risk, skin conductance is high risk, and coordination is medium risk during this period. The risk level interval coefficient refers to the distribution ratio or weight of each risk level within a certain time period. For example, it can be represented as a vector [low risk ratio, medium risk ratio, high risk ratio] by statistically analyzing the frequency of each risk level over a period of time, reflecting the overall distribution of the comprehensive indicator across each risk interval. This coefficient can be used for weight adjustment in subsequent indicator combination analysis and also facilitates the visualization and tracking of health status in knowledge graphs.

[0035] The indicator combination analysis submodule calls the level interval coefficient, compares the fluctuation direction of multiple feature values ​​within the same time period, analyzes the consistency of waveform trends, identifies the combination logic between multiple indicators, and obtains the indicator logic matching rate. In the indicator combination analysis submodule, the grade interval coefficient mainly serves as the basis for feature selection and weight adjustment. Specifically, when analyzing the consistency of the fluctuation direction of multiple features within the same time period, the system prioritizes features with higher grade interval coefficients (i.e., those frequently occurring in high-risk or medium-risk intervals), focusing on the consistency of the fluctuation trends of these features and assigning them higher weight or priority when calculating the logical matching rate. Although the basic calculation for obtaining the indicator logical matching rate mainly relies on the directional consistency of the first-order differences of each indicator, the grade interval coefficient can influence which features are included in the combination analysis and the interpretation and classification of the final matching rate, thus making the combination logical analysis results more focused on indicators with higher health risks or significant changes. The indicator combination analysis submodule calls the level interval coefficient. By comparing the fluctuation direction of multiple feature values ​​within the same time period, it analyzes the consistency of waveform trends and identifies the combination logic between multiple indicators. First, it takes the first-order difference sequence of continuous data sequences of indicators such as heart rate, blood oxygen, skin conductance, and skin temperature within the time window. By judging the sign of the difference between the current data and the previous data, it extracts the direction of change. For example, if the heart rate data sequence is 74, 76, 79, 77, 75, its first-order difference is +2, +3, -2, -2, so the direction of change is increase, increase, decrease, decrease. Simultaneously, it takes blood oxygen data of 96, 97, 98, 96, 94, so the direction of change is increase, increase, decrease, decrease. The consistency of the direction of change between the two within this window is 100%, that is, the number of times the direction is the same at each time point divided by the total length of 4, which gives a consistency rate of 1. A consistency rate above 0.8 is defined as a combined consistent event. After identifying events for all monitoring periods, the frequency of occurrence of combined consistent events of multiple indicators is statistically analyzed. The calculation formula is: ,in For the logical matching rate of the indicator, The number of events grouped in the same direction. Given the total number of calculation segments, for example, if 10 minutes is divided into 120 5-second windows, and the number of consistent events detected is 78, then... The ratio is then interval-classified: 0–0.3 is a weak match, 0.3–0.7 is a medium match, and 0.7–1 is a strong match. The matching level and ratio are written together into the structure field to obtain the logical match rate of the indicator.

[0036] The health profile construction submodule calls the indicator logic matching rate, analyzes the distribution of the level range of each feature value, evaluates the correlation level between multiple sets of indicators and defines the individualized node structure, constructs an individualized structured health profile, and establishes a health knowledge graph dataset. The health profile construction submodule calls the logical matching rate of indicators, analyzes the distribution of each feature value's level interval, evaluates the correlation hierarchy between multiple sets of indicators, and defines an individualized node structure to construct an individualized structured health profile. The process first filters all indicator combinations based on the logical matching rate output by the previous module, selects feature combinations in strong matching segments to establish node connections, and constructs node attribute values ​​according to the position of each indicator in the level interval. For example, if heart rate is at level 2, blood oxygen is at level 2, and skin temperature is at level 1, then this combination node is encoded as H2_O2_T1, forming a healthy state node. If this state repeats for three consecutive time windows, it is considered a stable healthy node, and this node is set as the core node and the starting point of the graph. Subsequently, the co-occurrence frequency between nodes is counted, an adjacency matrix is ​​constructed, and the correlation strength between indicators is determined based on the connection density, using the formula: ,in Let represent the correlation strength between the i-th indicator and the j-th indicator. The number of times two indicators co-occur in the same segment. Let be the total number of times the i-th indicator appears. If the number of times heart rate and blood oxygen co-occurs is 48, and heart rate appears a total of 60 times, then... When the strength is greater than 0.6, it is set as a real connection, and the others are virtual connections. Finally, a node-edge-weight triplet structure is constructed, and the output is the user's individual health graph information, and a health knowledge graph dataset is established.

[0037] The trend modeling module includes: The behavior recognition submodule is based on a health knowledge graph dataset. It calls continuous time series of acceleration data and skin conductance data to calculate the user's activity intensity in real time, detect fall events, and generate activity status values. In the calculation of user activity intensity, the health knowledge graph dataset primarily serves as background support and intelligent filtering. The sensitivity or reference standard for activity intensity calculation can be dynamically adjusted based on the user's health node attributes (such as past heart rate fluctuations, skin conductance characteristics, risk levels, etc.). For example, a more sensitive activity intensity threshold can be set for high-risk users, or historical health characteristics can be combined to determine whether detected strong acceleration changes might be associated with actual falls. Furthermore, the knowledge graph can be used to personalize the interpretation of activity intensity results and associate them with health profiles, helping the system more accurately identify abnormal behaviors and generate activity status values. Although the formula itself does not directly call upon knowledge graph data, the knowledge graph provides crucial personalized support and contextual information in the parameter setting, risk assessment, and result interpretation stages of activity intensity calculation.

[0038] The specific formula for calculating the intensity of a user's activity in real time is as follows: ; Calculate the user's activity intensity value; in, For the first The user's activity intensity value at any given time. For the first The acceleration value in the X-axis direction at time 1. For the first The acceleration value in the Y-axis direction at time 1. For the first The acceleration value in the Z-axis direction at time 1. For the first skin electrical activity conductivity value at any time For the first skin electrical activity conductivity value at any time For the first The sampling time point at time . For the first The sampling time point before the previous time. The normalized coefficients for the acceleration component are... This is the normalization coefficient for the change in skin electrical activity.

[0039] formula: ; Detailed explanation of the formula and its calculation derivation: The formula is used to calculate the user's activity intensity value at a single moment, to judge the user's physical activity status in real time and to assist in fall event detection. The result is used to generate activity status value, which serves as the basis for subsequent health status trend analysis and risk assessment. Parameter meanings and settings: For the first The acceleration value along the X-axis at any given time, in meters per second squared, with a set detection value of 0.5 meters per second squared; For the first The acceleration value in the Y-axis direction at any given time, in meters per square second, with a set detection value of 0.4 meters per square second; For the first The acceleration value in the Z-axis direction at any given time, in meters per square second, with a set detection value of 0.7 meters per square second; For the first The constant-time skin conductance value, in microsiemens, with a set detection value of 0.52 microsiemens; For the first The constant-time skin conductance value, in microsiemens, with a set detection value of 0.48 microsiemens; For the first The sampling time point is set to 10 seconds. For the first The sampling time point is set to 9 seconds. This is the normalization coefficient for the acceleration component, used to adjust the weight of the numerical influence of the acceleration characteristics, and is set to 1. This is the normalization coefficient for the skin conductance variation component, used to adjust the numerical influence weight of the skin conductance variation characteristics, and is set to 30.

[0040] Substitute the parameters into the formula to calculate: ; ; ; The result 2.1487 indicates the user's activity intensity value at the 10th second. This value reflects the current level of the user's physical activity and will be used as an activity status value in subsequent fall event detection and health trend change analysis.

[0041] The load assessment submodule calls the activity status value and combines it with the time series of continuous heart rate data to extract the heart rate variation amplitude and frequency corresponding to various activity states, assess the user's cardiac load status in real time, and obtain the cardiac load distribution rate. The load assessment submodule calls the activity status value and combines it with the time series of continuous heart rate data to extract the heart rate variation amplitude and frequency corresponding to various activity states. It assesses the user's cardiac load status in real time, dividing the heart rate data into sliding window segments, each with a time length of 60 seconds. Based on the activity type marked by the activity status value, it classifies the state into three categories: resting, walking, and strenuous exercise. For each state, it calculates the difference between the maximum and minimum heart rate values ​​to determine the heart rate variation amplitude. Taking the walking state as an example, if the maximum heart rate in this segment is 96 beats per minute and the minimum is 84 beats per minute, the amplitude is 12 beats per minute. The method for calculating the heart rate variation frequency is based on the difference between adjacent heart rate changes within this segment. If the number of times per minute exceeds 3, and there are 5 instances of changes exceeding the threshold within this period, then the frequency of change is recorded as 5. For different activity states, a cardiac load grading standard is set: a resting heart rate amplitude of less than 5 beats per minute is considered low load, 5–10 beats per minute is considered medium load, and more than 10 beats per minute is considered high load. The statistical heart rate amplitude and frequency under each state are classified into load levels, and finally the cardiac load distribution rate is calculated. The calculation method is to divide the duration of each load level by the total duration. If the low load state is 3 minutes, the medium load state is 5 minutes, and the high load state is 2 minutes within a 10-minute monitoring period, then the cardiac load distribution rates are 0.3, 0.5, and 0.2, respectively.

[0042] The trend extraction submodule, based on the cardiac load distribution rate, compares the user's activity intensity and cardiac load to assess and analyze the health change characteristics over a continuous period of time in real time, and establishes a health status trend value. The trend extraction submodule, based on the cardiac load distribution rate, assesses and analyzes health change characteristics over continuous time periods in real time by comparing the user's activity intensity and cardiac load. During execution, it first retrieves pre-calculated activity intensity data and cardiac load distribution rate data to establish a time-axis alignment. It then compares the average activity intensity with the cardiac load level within each time period to determine if there are cases where the cardiac load level is higher than expected due to activity intensity. The criterion is that when the activity intensity is below 0.5 m / s² and the cardiac load is high, it is considered an abnormal state. Continuous time periods are marked, and the duration of consecutive abnormal states is accumulated. If the accumulated duration exceeds 3 minutes, it is recorded as a high-risk trend in the health change characteristics. Statistical analysis is performed on the comparison between activity intensity and cardiac load across all monitoring periods, outputting indicators such as the number of abnormal trends, the duration of consecutive occurrences, and the maximum magnitude of cardiac load change within the monitoring period. These characteristic data are integrated into structured data fields for health change characteristics, establishing a health status trend value. The health status trend value is a structured comprehensive indicator reflecting the dynamic relationship between the user's physiological activity and cardiac load, the characteristics of abnormal states, and their risk changes within a certain monitoring period. Its specific content includes: comparing the average activity intensity with the cardiac load level within each consecutive time period, identifying and accumulating the duration, frequency, and maximum consecutive abnormal duration of abnormal trends in activity intensity and cardiac load mismatch (e.g., low activity intensity but high cardiac load), as well as the maximum amplitude of cardiac load changes throughout the entire monitoring period. The health status trend value integrates these analysis results to form data fields including the number of abnormal trend segments, the duration of each segment, the maximum amplitude of load changes, and the trend risk level. This data is used to dynamically reflect the user's recent health changes and provides key basis for individualized health risk assessment, dynamic threshold adjustment, and subsequent intervention decisions.

[0043] The rule-based judgment module includes: The threshold adjustment submodule is based on the health status trend value, calls the distribution range of individual physiological parameters in the structured health profile, calculates the distribution offset of blood oxygen value and respiratory rate value in multiple level intervals under the current user status, adjusts the blood oxygen detection threshold and respiratory rate abnormal judgment threshold according to the offset direction, and generates an individualized monitoring threshold group. In the calculation of distribution offset, the health status trend value primarily serves as dynamic background information on changes in key physiological parameters such as blood oxygen and respiratory rate, helping to interpret the risk significance of the current distribution offset result. Although the calculation formula for distribution offset itself only involves comparing blood oxygen and respiratory rate with the health profile interval for the current monitoring period, the health status trend value can provide contextual support for threshold adjustment and risk interpretation. For example, when the health status trend value shows a high-risk trend, even if the distribution offset value is small, the system can still prompt for greater caution, appropriately tightening individualized thresholds or increasing the monitoring alert level; conversely, if the trend value is stable, a certain degree of fluctuation can be tolerated. Thus, although the trend value does not directly participate in the formula calculation, it plays a regulatory and interpretive role in subsequent threshold generation, early warning logic, and individualized judgment, making the practical application of distribution offset more closely aligned with the user's dynamic health status.

[0044] The specific formula for calculating the distribution offset of a user's current blood oxygen level and respiratory rate across multiple level intervals is as follows: ; Calculate the offset equalization value; in, This represents the joint offset equilibrium value of the distribution of blood oxygen and respiratory rate within the current user's healthy range. This represents the average blood oxygen level during the current monitoring period. This represents the median value of an individual's blood oxygen saturation range recorded in the health profile. Represents the upper limit of an individual's blood oxygenation range. Represents the lower limit of an individual's blood oxygenation range. This represents the average respiratory rate during the current monitoring period. This represents the median value of the individual's respiratory rate range recorded in the health profile. Represents the upper limit of an individual's respiratory rate range. This represents the lower limit of an individual's respiratory rate range.

[0045] formula: ; Detailed explanation of the formula and its calculation derivation: The formula is used to calculate the joint distribution offset balance value of the user's blood oxygen value and respiratory rate value in the individual health standard range under the current state. The result is used to adjust the blood oxygen detection threshold and the respiratory rate abnormality judgment threshold. Parameter meanings and settings: This is the average value of the user's blood oxygen data within the current monitoring period, expressed as a percentage. The detection value is set at 96% based on the continuous blood oxygen sampling data from the wristband device during a 5-minute monitoring period. The median blood oxygen saturation range for an individual in the structured health profile, expressed as a percentage, is set at 95.5%. The upper limit of an individual's blood oxygen saturation range in a structured health profile, expressed as a percentage, is set to 100%. The lower limit of an individual's blood oxygen saturation range in a structured health profile, expressed as a percentage, is set at 92%. This is the average respiratory rate of the user within the current monitoring period, expressed in breaths per minute. Based on the continuous respiratory rate sampling data of the wristband device during a 5-minute monitoring period, it is set to 18 breaths per minute. The median respiratory rate range for an individual in the structured health profile, in breaths per minute, is set to 16 breaths per minute. The upper limit of the individual's respiratory rate range in the structured health profile, in units of breaths per minute, is set to 20 breaths per minute; The lower limit of the individual's respiratory rate range in the structured health profile, in units of breaths per minute, is set to 12 breaths per minute.

[0046] Substitute the parameters into the formula to calculate: ; ; ; The result of 0.3125 indicates the degree of joint deviation between the user's blood oxygen value and respiratory rate value within the individual health standard range during the monitoring period. The larger the value, the more serious the deviation of the user's current state from the individual health standard range. This result is used for subsequent dynamic adjustment of the blood oxygen detection threshold and the respiratory rate abnormality judgment threshold. Based on the direction and magnitude of the deviation, the settings of the blood oxygen detection threshold and the respiratory rate abnormality judgment threshold are corrected to generate an individualized monitoring threshold group.

[0047] When adjusting the anomaly detection threshold based on the offset balance value, the system automatically adjusts the threshold sensitivity according to the deviation of the current blood oxygen and respiratory rate from their respective healthy standard ranges. When the distribution offset increases significantly, indicating that the user's physiological state is closer to the risk edge, the system will correspondingly raise the minimum warning line for blood oxygen or lower the maximum warning line for respiratory rate, making anomaly detection more stringent. When the offset is small, indicating that the user's state is relatively stable, the system can appropriately relax the range of anomaly detection to avoid false alarms. The direction of the offset is determined by comparing the current average value with the median value of the healthy range: if the current average blood oxygen value is lower than the healthy median value, it is considered downward skewed, and the focus is mainly on adjusting the lower limit of blood oxygen; if the average respiratory rate is higher than the healthy median value, it is considered upward skewed, and the focus is mainly on adjusting the upper limit of respiratory rate. This allows for targeted adjustment of the corresponding threshold based on the direction of the offset, ensuring that anomaly detection is both sensitive and individualized.

[0048] The anomaly extraction submodule calls the individualized monitoring threshold group to compare real-time blood oxygen data and respiratory rate data, detect abnormal data and locate the time period of the abnormal state, and obtain the abnormal state interval value. The anomaly extraction submodule calls the individualized monitoring threshold group to compare real-time blood oxygen data and respiratory rate data, detect abnormal data, and locate the time period of the abnormal state. During the operation, it first iterates through the real-time blood oxygen data for the current time period, acquiring sampled values ​​second by second, and then compares each value with the lower limit threshold of blood oxygen in the individualized monitoring threshold group. When a blood oxygen value is detected to be lower than the set threshold, this time point is recorded as an anomaly detection point. The duration of consecutive anomaly detection points is accumulated, and when the accumulated duration exceeds the set minimum anomaly duration threshold of 5 seconds, it is marked as an anomaly. For abnormal time periods, assuming that the blood oxygen value is continuously below 90.25% from 12:00:10 to 12:00:20, this period is recorded as an abnormal time period. Similarly, respiratory rate data is detected, and the respiratory rate sampling value per minute is judged. When the sampling value is greater than 19.6 times per minute, an abnormal detection point is recorded. If the continuous abnormal detection duration exceeds 5 seconds, it is positioned as an abnormal time period. All detected abnormal time periods are merged, and adjacent non-abnormal intervals less than 2 seconds are merged into the same segment to obtain the abnormal time period value.

[0049] The risk assessment submodule calculates the user's health risk level in real time based on the abnormal state range value and the characteristics of health changes, and establishes a health risk assessment value. The risk assessment submodule calculates the user's health risk level in real time based on the abnormal state interval values ​​and health change characteristics. During processing, it first accumulates the duration of each abnormal interval within the abnormal state interval values ​​to obtain the cumulative duration of the abnormal state. For example, assuming the user has three abnormal intervals within the monitoring period, with durations of 10 seconds, 15 seconds, and 20 seconds respectively, the cumulative duration is 45 seconds. Simultaneously, it calls upon the cardiac load change trend from the health change characteristics to calculate the maximum load level. If the maximum load level is high load, the cumulative abnormal duration is then risk-weighted, with a high load state weighting coefficient of 1.5, and the weighted abnormal duration is calculated as follows: The data is then divided according to the health risk level judgment criteria. The health risk level judgment rules are set as follows: a weighted abnormal duration of less than 30 seconds is low risk, 30 to 60 seconds is medium risk, and more than 60 seconds is high risk. Based on the calculation results, the weighted abnormal duration is 67.5 seconds, and the health risk level is judged to be high risk. Finally, the user's abnormal state duration, load status, and risk level are integrated to establish a health risk assessment value.

[0050] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the flow or function according to the embodiments of the present invention is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0051] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0052] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0053] It should be understood that, in various embodiments of the present invention, the sequence number of each process does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0054] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0055] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0056] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0057] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0058] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0059] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0060] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A health monitoring wristband based on multimodal data fusion and knowledge graph, characterized in that, The system includes: The physiological data acquisition module calls the wristband sensor to collect various physiological and acceleration data. Through timestamps, it performs time alignment and data interpolation on the data, identifies and calibrates abnormal fluctuation data segments, and generates a multi-source fusion dataset. Based on the multi-source fusion dataset, the feature extraction module extracts the fluctuation amplitude, fluctuation frequency, and stability duration of each physiological parameter, analyzes the synchronicity of heart rate and blood oxygen, assesses the coordination of skin conductance and temperature, and generates comprehensive feature index values. Based on the comprehensive feature index values ​​and combined with preset health standards, the knowledge graph construction module identifies health level ranges for multiple feature values. By analyzing and identifying the combination logic between multiple indicators, it constructs an individualized structured health profile and generates a health knowledge graph dataset. The trend modeling module, based on the health knowledge graph dataset, calls continuous time series of acceleration, skin conductance, and heart rate data to analyze the user's activity intensity and detect fall events. It combines the fluctuation frequency of heart rate data in various activity states to assess the user's cardiac load status, extracts the user's health change characteristics over continuous time periods, and obtains health status trend values.

2. The health monitoring bracelet based on multimodal data fusion and knowledge graph as described in claim 1, characterized in that, The multi-source fusion dataset includes the original sensor dataset, time-aligned records, and dataset interpolation results. The comprehensive feature index values ​​include the amplitude of physiological parameter changes, the synchronization degree of fluctuations between parameters, the duration of resting stability, and the degree of skin signal coupling. The health knowledge graph dataset specifically includes individual physiological parameter health level identifiers, cross-parameter index combination patterns, and structured health status nodes. The health status trend values ​​include the amplitude of activity intensity changes, fall event recognition status, and continuous change trajectory of health characteristics.

3. The health monitoring bracelet based on multimodal data fusion and knowledge graph as described in claim 1, characterized in that, The physiological data acquisition module includes: The data acquisition submodule calls the wristband sensor to acquire the user's heart rate data, blood oxygen data, skin temperature data, skin conductance data, and real-time acceleration data, and establishes the raw sensor dataset; The data alignment submodule, based on the original sensor dataset, calls the timestamp of each data item to perform time alignment on the data, and combines data interpolation to supplement the data for missing time periods, thereby establishing a time-aligned dataset. The data calibration submodule, based on the time-aligned dataset, calls the change trend of each data item, calculates the fluctuation amplitude of heart rate data, blood oxygen data, skin temperature data, and skin conductance data, identifies abnormal fluctuation data segments and performs calibration, and obtains a multi-source fusion dataset.

4. The health monitoring bracelet based on multimodal data fusion and knowledge graph according to claim 3, characterized in that, The feature extraction module includes: The data feature calculation submodule extracts the numerical variation range of each physiological parameter based on the multi-source fusion dataset, calculates the amplitude of the numerical change of each data in a continuous period, performs frequency statistics on the number of changes of data within a time window, obtains the stable duration of the numerical change of each data, and establishes basic feature parameter values. The feature association submodule calls the basic feature parameter values ​​to perform synchronization calculations on the changing trends of heart rate data and blood oxygen data, calculates the time difference of data change peaks, and obtains the physiological synchronization correlation degree. The feature statistics submodule evaluates the coordination of the changing trends of skin electrodermatology data and skin temperature data based on the physiological synchronization correlation, calculates the difference rate of data change magnitude, and establishes a comprehensive feature index value.

5. The health monitoring bracelet based on multimodal data fusion and knowledge graph according to claim 4, characterized in that, The knowledge graph construction module includes: The health interval identification submodule identifies health level intervals for multiple feature values ​​by comparing them with preset health standards based on the comprehensive feature index values, and establishes level interval coefficients. The indicator combination analysis submodule calls the level interval coefficient, compares the fluctuation direction of multiple feature values ​​within the same time period, analyzes the consistency of waveform trends, identifies the combination logic between multiple indicators, and obtains the indicator logic matching rate. The health profile construction submodule calls the logical matching rate of the indicators, analyzes the distribution of the level range of each feature value, evaluates the correlation level between multiple sets of indicators and defines the individualized node structure, constructs an individualized structured health profile, and establishes a health knowledge graph dataset.

6. The health monitoring bracelet based on multimodal data fusion and knowledge graph according to claim 5, characterized in that, The trend modeling module includes: The behavior recognition submodule, based on the health knowledge graph dataset, calls the continuous time series of acceleration data and skin conductance activity data to calculate the user's activity intensity in real time, detect fall events, and generate activity status values. The load assessment submodule calls the activity status value, combines it with the time series of continuous heart rate data, extracts the heart rate variation amplitude and frequency corresponding to various activity states, assesses the user's cardiac load status in real time, and obtains the cardiac load distribution rate. The trend extraction submodule, based on the cardiac load distribution rate, compares the user's activity intensity and cardiac load to assess and analyze the health change characteristics over a continuous period of time in real time, and establishes a health status trend value.

7. The health monitoring bracelet based on multimodal data fusion and knowledge graph according to claim 6, characterized in that, The specific formula for calculating the user's activity intensity in real time is as follows: ； Calculate the user's activity intensity value; in, For the first The user's activity intensity value at any given time. For the first The acceleration value in the X-axis direction at time 1. For the first The acceleration value in the Y-axis direction at time 1. For the first The acceleration value in the Z-axis direction at time 1. For the first skin electrical activity conductivity value at any time For the first skin electrical activity conductivity value at any time For the first The sampling time point at time . For the first The sampling time point before the previous time. The normalized coefficients for the acceleration component are... This is the normalization coefficient for the change in skin electrical activity.

8. The health monitoring bracelet based on multimodal data fusion and knowledge graph according to claim 1, characterized in that, The system also includes: Based on the health status trend value, the rule judgment module uses an individualized structured health profile to adjust the blood oxygen threshold and respiratory rate abnormality detection threshold, detects abnormal blood oxygen data and respiratory rate data in real time, extracts the time period of abnormal status, combines the user's health change characteristics, calculates the user's health risk level, and generates a health risk assessment value. The health risk assessment values ​​specifically refer to blood oxygen risk level, abnormal respiratory rate interval label, and abnormal distribution coefficient of health trend.

9. The health monitoring bracelet based on multimodal data fusion and knowledge graph according to claim 8, characterized in that, The rule determination module includes: The threshold adjustment submodule, based on the health status trend value, calls the distribution range of individual physiological parameters in the structured health profile, calculates the distribution offset of blood oxygen value and respiratory rate value in multiple level intervals under the current user status, adjusts the blood oxygen detection threshold and respiratory rate abnormal judgment threshold according to the offset direction, and generates an individualized monitoring threshold group. The anomaly extraction submodule calls the individualized monitoring threshold group to compare real-time blood oxygen data and respiratory rate data, detects abnormal data and locates the time period of the abnormal state, and obtains the abnormal state interval value. The risk assessment submodule calculates the user's health risk level in real time based on the abnormal state range value and the characteristics of health changes, and establishes a health risk assessment value.

10. The health monitoring bracelet based on multimodal data fusion and knowledge graph according to claim 9, characterized in that, The specific formula for calculating the distribution offset of the user's current blood oxygen value and respiratory rate value across multiple level intervals is as follows: ； Calculate the offset equalization value; in, This represents the joint offset equilibrium value of the distribution of blood oxygen and respiratory rate within the current user's healthy range. This represents the average blood oxygen level during the current monitoring period. This represents the median value of an individual's blood oxygen saturation range recorded in the health profile. Represents the upper limit of an individual's blood oxygenation range. Represents the lower limit of an individual's blood oxygenation range. This represents the average respiratory rate during the current monitoring period. This represents the median value of the individual's respiratory rate range recorded in the health profile. Represents the upper limit of an individual's respiratory rate range. This represents the lower limit of an individual's respiratory rate range.

Images