Health information management method, system and equipment based on motion data and medium

Abstract

The invention relates to a health information management method, system and device based on motion data and a medium. The method comprises the following steps: acquiring real-time exercise physiological data and living work and rest data of a user; based on the real-time exercise physiological data, analyzing the activity state of the user to obtain a real-time activity state label; based on the real-time exercise physiological data, the real-time activity state tag and the daily work and rest data, analyzing the characteristics of the body change of the user to obtain a comprehensive context characteristic vector; the comprehensive context feature vector comprises an instantaneous context feature vector, an activity period context feature vector and a work and rest rule context feature vector; generating a dynamic early warning threshold based on the real-time motion physiological data, the comprehensive context feature vector and a preset causal relationship graph; and based on the dynamic early warning threshold and the real-time exercise physiological data, interpretable early warning information is generated. By adopting the method, the accuracy and scientificity of risk early warning can be improved.

Description

Technical Field

[0001] This invention belongs to the field of health monitoring, and in particular relates to a method, system, device and medium for health information management based on exercise data. Background Technology

[0002] With the development of health monitoring technology, health risk warning technology based on exercise data has emerged. This technology achieves real-time health monitoring by continuously collecting users' physiological indicators (such as heart rate and exercise data), which leads to the widely used health warning method based on static thresholds.

[0003] Existing health alert methods based on static thresholds typically employ fixed threshold settings, such as a preset upper limit for heart rate (e.g., 120 bpm). When real-time heart rate data exceeds this threshold, an alarm is triggered directly. The process involves acquiring raw heart rate, accelerometer, and gyroscope data from sensors, performing simple filtering and noise reduction, directly comparing it with the preset static threshold, and generating an alert based on the comparison result. This approach is simple, efficient, and uses the same standard for all users and scenarios, making it easy to deploy and widely adopted.

[0004] However, existing health alert methods based on static thresholds do not differentiate between users' activity states (such as exercise or rest) or consider individual historical health trends or environmental context. They rely solely on a single numerical value to determine abnormalities, leading to a high false alarm rate. For example, a physiological increase in heart rate during normal exercise can frequently trigger false alarms, while subtle but dangerous abnormal changes at rest may be overlooked. This not only leads to user alarm fatigue and decreased trust but also masks real health risks, reducing the accuracy and scientific validity of risk alerts. Summary of the Invention

[0005] Therefore, it is necessary to provide a health information management method, system, device, and medium based on exercise data that can improve the accuracy and scientific validity of risk warning in response to the above-mentioned technical problems.

[0006] Firstly, this application provides a health information management method based on exercise data, including:

[0007] Acquire real-time exercise physiological data and daily routine data of the user; real-time exercise physiological data includes heart rate parameters, accelerometer parameters, and gyroscope parameters; daily routine data is used to characterize the user's daily routine.

[0008] Based on real-time motion physiological data, the user's activity status is analyzed to obtain real-time activity status labels; these labels are used to characterize the user's real-time activity type.

[0009] Based on real-time exercise physiological data, real-time activity status tags, and daily routine data, the characteristics of user's physical changes are analyzed to obtain a comprehensive contextual feature vector. The comprehensive contextual feature vector includes instantaneous contextual feature vector, activity cycle contextual feature vector, and daily routine contextual feature vector.

[0010] Based on real-time exercise physiological data, comprehensive contextual feature vectors, and a pre-defined causal relationship map, a dynamic warning threshold is generated; and based on the dynamic warning threshold and real-time exercise physiological data, interpretable warning information is generated; the interpretable warning information is used to characterize the health risks and causes of the user.

[0011] Furthermore, based on real-time motion physiological data, the user's activity status is analyzed to obtain real-time activity status labels, including:

[0012] Based on a preset physiological data threshold, real-time exercise physiological data is filtered to obtain filtered exercise physiological data.

[0013] Based on a preset detection time window, the filtered motion physiological data is segmented to obtain the motion physiological data to be tested; the motion physiological data to be tested includes the heart rate parameters, accelerometer parameters, and gyroscope parameters to be tested.

[0014] Based on the accelerometer parameters to be measured, the user's activity intensity is quantified to obtain activity intensity parameters;

[0015] Calculate the variance of the parameters of the gyroscope under test to obtain the gyroscope variance parameters; calculate the variance of the parameters of the accelerometer under test to obtain the acceleration variance parameters.

[0016] The exercise physiological data to be tested is transformed in the frequency domain to obtain the exercise physiological spectrum; and the spectral entropy of the exercise physiological spectrum is calculated to obtain the data complexity parameter.

[0017] The physiological spectrum of the motion to be measured, the gyroscope variance parameter, the acceleration variance parameter, and the data complexity parameter are input into the preset activity state classification model to obtain the real-time activity state label; the activity state classification model is constructed based on the random forest classifier.

[0018] Furthermore, based on real-time exercise physiological data, real-time activity status tags, and daily routine data, the characteristics of user physical changes are analyzed to obtain a comprehensive contextual feature vector, including:

[0019] Based on the preset instantaneous time window and the preset instantaneous time decay factor, the mean and variance of heart rate parameters are calculated to obtain the instantaneous context feature vector.

[0020] Based on real-time activity status labels and preset periodic time windows, instantaneous context feature vectors are selected from preset historical motion physiological data to form a periodic instantaneous context feature vector set; historical motion physiological data includes real-time activity status labels of users' historical motions and corresponding instantaneous context feature vectors;

[0021] Based on the preset cycle time decay factor, the mean and variance of the instantaneous context feature vectors in the cycle instantaneous context feature vector set are calculated to obtain the activity cycle context feature vector.

[0022] Based on daily routine data, we analyze the changing trends of users' daily routines to obtain contextual feature vectors of daily routines.

[0023] A comprehensive contextual feature vector is obtained based on the instantaneous contextual feature vector, the activity cycle contextual feature vector, and the work-rest pattern contextual feature vector.

[0024] Furthermore, the daily routine data includes at least the parameters of sleep duration, deep sleep ratio, number of times the person wakes up, average steps on the first day and average steps on the second day; the sampling timestamp of the average steps on the first day is later than the sampling timestamp of the average steps on the second day.

[0025] Based on daily routine data, we analyze the changing trends of users' daily routines to obtain contextual feature vectors of daily routines, including:

[0026] Based on a pre-set sleep parameter database, the daily routine data is normalized to obtain normalized daily routine data; the normalized daily routine data includes normalized sleep duration parameters, normalized deep sleep ratio parameters, and normalized number of awakenings.

[0027] Based on normalized daily routine data, sleep quality parameters are calculated using the following formula:

[0028]

[0029] in, It is a sleep quality parameter. It's a time-based weighting. It is a normalized sleep duration parameter. It is deep sleep weight. It is the normalized deep sleep ratio parameter. It's the weight of waking up. It is the normalized number of times you wake up;

[0030] Based on the average steps taken on the first day and the average steps taken on the second day, the activity trend parameter is calculated using the following formula:

[0031]

[0032] in, It is an activity level trend parameter. It is the average number of steps taken on the first day. This is the average number of steps taken on the second day;

[0033] Based on sleep quality parameters and activity trend parameters, a contextual feature vector of daily routine patterns is obtained.

[0034] Furthermore, the causal relationship map was obtained through the following method:

[0035] Obtain a preset historical physiological database; the historical physiological database includes the user's historical sleep quality parameters, average daily steps, and heart rate parameters with real-time activity status tags;

[0036] Using heart rate parameters, sleep quality parameters, and average daily steps as nodes, an initial causal relationship graph was constructed.

[0037] Based on the initial causal relationship graph, the independence of each node in the initial causal relationship graph is verified to obtain the correlation coefficient between each node; and based on the correlation coefficient, the connection edges between each node are constructed to obtain the complete causal relationship graph.

[0038] For each connecting edge in the complete causal relationship graph, the causal relationship between the nodes at both ends of the connecting edge is analyzed to obtain the causal relationship parameters. The causal relationship parameters include the causal relationship pointing value and the relationship strength. The causal relationship pointing value is used to indicate the causal relationship between nodes.

[0039] The causal relationship map is obtained by adjusting and improving the causal relationship parameters.

[0040] Furthermore, based on real-time exercise physiological data, comprehensive contextual feature vectors, and a pre-defined causal relationship map, a dynamic warning threshold is generated; and based on the dynamic warning threshold and real-time exercise physiological data, interpretable warning information is generated, including:

[0041] The basic warning threshold is calculated based on the activity cycle context feature vector, the instantaneous context feature vector, and the preset scaling factor.

[0042] Based on the causal relationship graph, nodes and causal relationship parameters affecting heart rate parameters are extracted to obtain a real-time quantitative set of influence relationships;

[0043] Based on the real-time quantitative impact relationship set and the preset historical parameter mean, the impact of the context feature vector of work and rest pattern on the basic early warning threshold is quantified to obtain the total compensation value of causal effect;

[0044] Based on the total compensation value of causal effects and the basic early warning threshold, the dynamic early warning threshold is calculated using the following formula:

[0045]

[0046] in, It is a dynamic early warning threshold. This is the basic early warning threshold. It is an adjustment factor. It is the total compensation value of causal effects;

[0047] The dynamic warning threshold and real-time motion physiological data are compared to obtain the physiological parameter comparison results; when the physiological parameter comparison results meet the preset warning requirements, a warning sign is obtained.

[0048] Based on early warning indicators, real-time quantitative impact relationship sets, and total causal effect compensation values, the contribution of the contextual feature vector of daily routines to the total causal effect compensation value is analyzed to obtain contribution ratio parameters. Based on dynamic early warning thresholds, physiological parameter comparison results, and contribution ratio parameters, interpretable early warning information is generated.

[0049] Secondly, this application also provides a health information management system based on exercise data, including:

[0050] The data acquisition module is used to acquire the user's real-time exercise physiological data and daily routine data; the real-time exercise physiological data includes heart rate parameters, accelerometer parameters, and gyroscope parameters; the daily routine data is used to characterize the user's daily routine.

[0051] The status analysis module is used to analyze the user's activity status based on real-time motion physiological data and obtain real-time activity status labels; the real-time activity status labels are used to characterize the user's real-time activity type.

[0052] The state feature extraction module is used to analyze the characteristics of user's physical changes based on real-time exercise physiological data, real-time activity status labels and daily routine data, and obtain a comprehensive contextual feature vector. The comprehensive contextual feature vector includes instantaneous contextual feature vector, activity cycle contextual feature vector and daily routine contextual feature vector.

[0053] The early warning information generation module is used to generate dynamic early warning thresholds based on real-time exercise physiological data, comprehensive contextual feature vectors, and preset causal relationship maps; and to generate interpretable early warning information based on dynamic early warning thresholds and real-time exercise physiological data; the interpretable early warning information is used to characterize the health risks and causes of the user.

[0054] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the health information management methods based on exercise data described in the first aspect of this application.

[0055] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the health information management methods based on motion data described in the first aspect of this application.

[0056] The aforementioned health information management method, system, device, and medium based on exercise data acquires the user's real-time exercise physiological data and daily routine data. Real-time exercise physiological data includes heart rate parameters, accelerometer parameters, and gyroscope parameters. Daily routine data characterizes the user's daily routine. Based on the real-time exercise physiological data, the system analyzes the user's activity state to obtain a real-time activity state label. This label characterizes the user's real-time activity type. Based on the real-time exercise physiological data, real-time activity state labels, and daily routine data, the system analyzes the characteristics of changes in the user's body to obtain a comprehensive contextual feature vector. This vector includes an instantaneous contextual feature vector, an activity cycle contextual feature vector, and a daily routine pattern contextual feature vector. Based on the real-time exercise physiological data, the comprehensive contextual feature vector, and a preset causal relationship graph, a dynamic warning threshold is generated. Based on the dynamic warning threshold and the real-time exercise physiological data, interpretable warning information is generated. This interpretable warning information characterizes the user's existing health risks and their causes. It achieves dynamic threshold adjustment based on individual physiological characteristics and activity context, overcoming the problem of high false alarm rate of traditional static threshold method under different activity states, improving the accuracy and personalization of health risk warning, and enhancing user trust and compliance through interpretable warning information. Attached Figure Description

[0057] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0058] Figure 1 A flowchart illustrating a health information management method based on exercise data, provided as an embodiment of this application;

[0059] Figure 2 This application provides a schematic diagram of the structure of a health information management system based on exercise data, as one embodiment of the present application.

[0060] Figure 3 This is a schematic diagram of the structure of a computer device for a health information management method based on exercise data, provided as an embodiment of this application. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0062] In one embodiment, such as Figure 1 As shown, a health information management method based on exercise data is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps S101-S104, wherein:

[0063] S101, acquire the user's real-time exercise physiological data and daily routine data; the real-time exercise physiological data includes heart rate parameters, accelerometer parameters and gyroscope parameters; the daily routine data is used to characterize the user's daily routine.

[0064] Specifically, the terminal acquires the user's real-time motion physiological data and daily routine data. The real-time motion physiological data includes heart rate parameters, accelerometer parameters, and gyroscope parameters; all these parameters have sampling timestamps and have undergone time alignment processing. The heart rate parameter, used to characterize the user's real-time heart rate, can be acquired through an optical heart rate sensor. The accelerometer parameters include accelerations in three axes: horizontal, vertical, and height, representing the user's real-time speed. The gyroscope parameters, also including horizontal, vertical, and height parameters, represent the user's real-time direction of movement. The daily routine data characterizes the user's daily routine, including but not limited to sleep duration and average daily steps. For example, the mathematical form of real-time motion physiological data can be: ,in It is real-time motion physiological data. It is the total number of sampling timestamps for heart rate parameters, accelerometer parameters, and gyroscope parameters. It is any sampling timestamp. The sampling timestamp is Heart rate parameters. The sampling timestamp is Horizontal acceleration. The sampling timestamp is The vertical acceleration. The sampling timestamp is The acceleration at height. The sampling timestamp is Horizontal gyroscope parameters. The sampling timestamp is The parameters of the vertical gyroscope. The sampling timestamp is Altitude gyroscope parameters.

[0065] S102, Based on real-time motion physiological data, analyze the user's activity status to obtain real-time activity status labels; real-time activity status labels are used to characterize the user's real-time activity type.

[0066] Specifically, the terminal performs multi-dimensional feature extraction and classification analysis on heart rate, accelerometer, and gyroscope parameters from real-time motion physiological data. By calculating the changing characteristics of these parameters, it analyzes the user's real-time activity state to obtain a real-time activity state label. This label characterizes the user's real-time activity type. For example, the real-time activity state label can be an activity state such as resting, walking, running, cycling, or sleeping.

[0067] S103 analyzes the characteristics of user's physical changes based on real-time exercise physiological data, real-time activity status tags, and daily routine data to obtain a comprehensive contextual feature vector. The comprehensive contextual feature vector includes instantaneous contextual feature vector, activity cycle contextual feature vector, and daily routine pattern contextual feature vector.

[0068] Specifically, the terminal analyzes the user's real-time exercise physiological data and real-time activity status tags at three time scales, obtaining instantaneous context feature vectors, activity cycle context feature vectors, and daily routine context feature vectors. The terminal then concatenates these three vectors to obtain a comprehensive context feature vector. The instantaneous context feature vector characterizes the changes in the current user's heart rate parameters within a preset instantaneous time window. The activity cycle context feature vector characterizes the changes in the current user's heart rate parameters within a preset cycle time window. The daily routine context feature vector represents the changing trend of the user's daily routine data. The preset instantaneous time window can be set according to the total number of heart rate parameter sampling timestamps and the instantaneous requirements in actual work; the default setting is 30 seconds. This embodiment does not further limit the setting of this preset instantaneous time window.

[0069] S104 generates dynamic warning thresholds based on real-time exercise physiological data, comprehensive contextual feature vectors, and preset causal relationship maps; and generates interpretable warning information based on dynamic warning thresholds and real-time exercise physiological data; the interpretable warning information is used to characterize the health risks and causes of the user.

[0070] Specifically, a pre-defined causal relationship graph is used to characterize the causal relationship and strength between lifestyle data and real-time exercise physiological data rate parameters. The terminal calculates a basic warning threshold for heart rate parameters based on the instantaneous context feature vector and activity cycle context feature vector in the comprehensive context feature vector. It then matches the lifestyle pattern context feature vector in the comprehensive context feature vector with the causal relationship and strength between the lifestyle data and real-time exercise physiological data rate parameters as represented in the pre-defined causal relationship graph, obtaining a total causal effect compensation value representing the impact of current lifestyle data on the current heart rate parameter threshold. This total causal effect compensation value is combined with the basic warning threshold to obtain a dynamic warning threshold. The terminal compares the real-time exercise physiological data rate parameters with the dynamic warning threshold. When the heart rate parameter exceeds the dynamic warning threshold, a warning is generated. A causal relationship analysis is performed on the total causal effect compensation value, analyzing the contribution of each dimension of the lifestyle pattern context feature vector to the current total causal effect compensation value, and generating interpretable warning information. Explainable warning information includes the contribution of each dimension of the feature vector of the daily routine context to the current total compensation value of the causal effect, the dynamic warning threshold, and the difference between the heart rate parameter and the dynamic warning threshold. It can alert users or medical professionals to the existence of risks and indicate which parameter in the daily routine data is responsible for the risk. For example, explainable warning information makes the warning logically transparent and clinically actionable by associating the warning decision (such as excessive heart rate) with multi-granular context (such as exercise) and upstream causes identified in the causal graph (such as recent decline in sleep quality).

[0071] The health information management method based on exercise data provided in this embodiment achieves dynamic threshold adjustment based on individual physiological characteristics and activity context through multi-scale temporal feature analysis and causal relationship reasoning. This overcomes the problem of high false alarm rate in traditional static threshold methods under different activity states, improves the accuracy and personalization of health risk warnings, and enhances user trust and compliance through interpretable warning information.

[0072] In one embodiment, based on real-time motion physiological data, the user's activity state is analyzed to obtain a real-time activity state label, including:

[0073] S201, based on a preset physiological data threshold, filters the real-time exercise physiological data to obtain filtered exercise physiological data.

[0074] Specifically, the terminal uses a Butterworth filter to perform low-pass filtering on the heart rate parameters, accelerometer parameters, and gyroscope parameters of the exercise physiological data based on preset physiological data thresholds, obtaining filtered heart rate parameters, filtered accelerometer parameters, and filtered gyroscope parameters. The terminal then concatenates these filtered parameters to obtain filtered exercise physiological data. The Butterworth low-pass filter is a filter with a frequency response as flat as possible within the passband, effectively removing high-frequency noise while preserving the inherent characteristics of the physiological signal. For example, the preset physiological data thresholds can be set according to the abnormal frequencies of the heart rate, accelerometer, and gyroscope parameters during actual operation. For the heart rate parameter, the physiological data threshold is set to 4Hz by default; for the accelerometer and gyroscope parameters, the physiological data threshold is set to 10Hz by default. This embodiment does not further limit the setting of these preset physiological data thresholds.

[0075] S202, based on a preset detection time window, segments the filtered motion physiological data to obtain the motion physiological data to be measured; the motion physiological data to be measured includes the heart rate parameter to be measured, the accelerometer parameter to be measured, and the gyroscope parameter to be measured.

[0076] Specifically, the terminal divides the filtered motion physiological data into sliding windows based on a preset detection time window, with the data within each window constituting a set of motion physiological data to be tested. ,in It refers to the heart rate parameter to be measured, including the heart rate parameter within the detection time window; These are the parameters of the accelerometer to be measured, including the accelerometer parameters for the three axes within the detection time window. The format can be: ,in It is the latest sampling timestamp in the filtered accelerometer parameters. It is a preset detection time window. It can be any sampling timestamp; These are the parameters of the accelerometer to be tested, including the gyroscope parameters within the detection time window. For example, the preset detection time window can be set based on the total number of sampling timestamps in actual work, with a default setting of 1 minute. This embodiment does not further limit the setting of this preset detection time window.

[0077] S203 quantifies the user's activity intensity based on the parameters of the accelerometer under test, and obtains the activity intensity parameters.

[0078] Specifically, the terminal calculates the signal amplitude area (SMA) based on the parameters of the accelerometer under test as an activity intensity parameter to quantify the user's activity intensity. The signal amplitude area (SMA) is a commonly used quantification of motion intensity, reflecting the overall motion intensity by calculating the sum of the absolute values ​​of the acceleration signals. The formula is as follows: ,in This is the number of sampling points within the detection time window. A higher SMA value indicates a higher activity intensity for the user. This parameter can effectively distinguish between activities of different intensities, such as standing still, walking, and running.

[0079] S204, calculate the variance of the parameters of the gyroscope under test to obtain the gyroscope variance parameters; calculate the variance of the parameters of the accelerometer under test to obtain the acceleration variance parameters.

[0080] Specifically, the terminal calculates the variances of the parameters of the gyroscope and the accelerometer under test, respectively, to obtain the gyroscope variance parameters. and acceleration variance parameter Variance calculation reflects the degree of fluctuation of sensor readings over time. The gyroscope variance parameter can capture the intensity of body rotation, while the acceleration variance parameter reflects the variability of linear motion.

[0081] S205, the exercise physiological data to be tested is transformed in the frequency domain to obtain the exercise physiological spectrum to be tested; and the spectral entropy of the exercise physiological spectrum to be tested is calculated to obtain the data complexity parameter.

[0082] Specifically, the terminal performs a Fast Fourier Transform on the exercise physiological data to be measured, converting the time-domain signal into a frequency-domain representation to obtain the exercise physiological spectrum to be measured; and uses the following formula: The spectral entropy of the calculated spectrum is obtained, and this spectral entropy is determined as a data complexity parameter. In the aforementioned formula, The physiological spectrum of the exercise to be measured is in frequency The normalized value of the power spectral density at that location.

[0083] S206, the measured motion physiological spectrum, gyroscope variance parameter, acceleration variance parameter and data complexity parameter are input into the preset activity state classification model to obtain the real-time activity state label; the activity state classification model is constructed based on the random forest classifier.

[0084] Specifically, the terminal inputs the measured motion physiological spectrum, gyroscope variance parameter, acceleration variance parameter, and data complexity parameter into a preset activity state classification model to obtain real-time activity state labels. The activity state classification model is built based on a random forest classifier. The random forest classifier is an ensemble learning method that improves classification accuracy and robustness by constructing multiple decision trees and combining their prediction results. The terminal uses preset historical motion data, trained with a random forest classifier, to input the current user's measured motion physiological spectrum, gyroscope variance parameter, acceleration variance parameter, and data complexity parameter, and outputs the corresponding activity state label in real time, such as "stationary," "walking," "running," "cycling," and "sleeping." The preset historical motion data includes a large amount of measured motion physiological spectrum, gyroscope variance parameter, acceleration variance parameter, and data complexity parameter labeled with activity state labels. This data can be collected according to actual work needs; this embodiment does not further limit the setting of this preset historical motion data.

[0085] The health information management method based on motion data provided in this embodiment achieves high-precision activity state recognition through multi-dimensional feature extraction and machine learning classification, providing accurate activity information for subsequent context-aware health early warning, overcoming the limitations of single feature recognition methods, and improving the robustness and adaptability of activity state recognition.

[0086] In one embodiment, based on real-time motion physiological data, real-time activity status tags, and daily routine data, the characteristics of changes in the user's body are analyzed to obtain a comprehensive contextual feature vector, including:

[0087] S301 calculates the mean and variance of heart rate parameters based on a preset instantaneous time window and a preset instantaneous time decay factor, and obtains the instantaneous context feature vector.

[0088] Specifically, the terminal segments the real-time exercise physiological data rate parameters based on a preset instantaneous time window to obtain an instantaneous heart rate parameter sequence. Then, based on the instantaneous heart rate parameter sequence and the instantaneous time decay factor, it calculates the short-term mean and variance of the heart rate parameters using an exponentially weighted moving average method. The exponentially weighted moving average is a time series smoothing method that assigns higher weight to recent data, enabling rapid response to changes in physiological state. The terminal uses the exponentially weighted moving average method with the following formula: and Iterative calculations are performed to obtain the mean and variance of the heart rate parameters at the latest sampling timestamp for the current user. These mean and variance values ​​are then concatenated to obtain the instantaneous context feature vector. It is any sampling timestamp in the instantaneous heart rate parameter sequence. ,in This is the latest sampling timestamp. It is a preset instantaneous time window. It is a sampling timestamp The mean over time. It is the instantaneous decay factor, used to control the decay rate of historical data. It is a sampling timestamp The mean over time. It is a sampling timestamp Variance over time. It is a sampling timestamp Heart rate parameters at that time. Sampling timestamp The variance over time. Its initial settings. and The value is 0. For example, the preset instantaneous time window can be set according to the total number of sampling timestamps of the heart rate parameters and the instantaneous requirements in actual work; the default setting is 30 seconds. This embodiment does not further limit the setting of this preset instantaneous time window. The preset instantaneous time decay factor can be set according to the requirements for controlling the decay of historical data in actual work; the default setting is 0.2. This embodiment does not further limit the setting of this preset instantaneous time decay factor. Optionally, the obtained instantaneous context feature vector is in the form of... .in, This is the latest sampling timestamp. The sampling timestamp is The instantaneous context feature vector. It is a sampling timestamp The mean over time. It is a sampling timestamp Variance over time.

[0089] S302, based on real-time activity status labels and preset periodic time windows, selects instantaneous context feature vectors from preset historical motion physiological data to form a periodic instantaneous context feature vector set; historical motion physiological data includes real-time activity status labels of the user's historical motion and corresponding instantaneous context feature vectors.

[0090] Specifically, the preset historical motion physiological data includes the user's real-time activity status labels and corresponding instantaneous context feature vectors, which can take the following forms: ,in, It is preset historical exercise physiological data. It is any real-time activity status tag. Each This refers to the user's past activities, specifically the real-time activity status tag. And the sampling timestamp is The instantaneous context feature vector is obtained by collecting historical motion physiological data of the user during historical activities and corresponding real-time activity status tags. The terminal selects instantaneous context feature vectors from the preset historical motion physiological data whose real-time activity status tags satisfy the current real-time activity status tag and whose sampling timestamp is within a preset periodic time window, forming a periodic instantaneous context feature vector set. For example, the preset periodic time window can be set according to the periodic requirements in actual work, with a default setting of two weeks. This embodiment does not further limit the setting of this preset periodic time window. Optionally, the form of the periodic instantaneous context feature vector set can be... ,in It is a set of periodic instantaneous context feature vectors. This is the latest sampling timestamp. It is a preset periodic time window. It is any sampling timestamp. The real-time activity status label is the current real-time activity status label. And the sampling timestamp is The instantaneous context feature vector. Each of these features... .in, It is a sampling timestamp The mean over time. It is a sampling timestamp Variance over time.

[0091] S303, based on the preset cycle time decay factor, calculates the mean and variance of the instantaneous context feature vectors in the cycle instantaneous context feature vector set, and obtains the activity cycle context feature vector.

[0092] Specifically, the terminal calculates the short-term mean and variance of heart rate parameters using an exponentially weighted moving average method based on a preset periodic decay factor. Specifically, the terminal iteratively calculates the mean and variance of the heart rate parameters within the preset periodic time window using an exponentially weighted moving average method for each sampling timestamp in the periodic instantaneous context feature vector set. The mean and variance of the aforementioned heart rate parameters are then concatenated to obtain the instantaneous context feature vector. For example, the preset periodic decay factor can be set according to the periodic decay pattern of the heart rate parameters and actual working conditions; the default setting is 0.02. This embodiment does not further limit the setting of this preset periodic decay factor. Optionally, the form of the activity periodic context feature vector can be... ,in, It is the activity cycle context feature vector. It is the average heart rate parameter in the current cycle. It is the variance of the heart rate parameters in the current cycle.

[0093] S304, based on daily routine data, analyzes the changing trends of users' daily routines to obtain the contextual feature vector of daily routines.

[0094] Specifically, the terminal analyzes the user's long-term sleep patterns based on sleep duration and average daily steps in daily routine data to obtain a contextual feature vector of the sleep pattern. For example, the terminal can quantify the user's sleep status based on sleep duration to obtain sleep quality parameters; and analyze the user's activity trends through average daily steps to obtain activity trend parameters. The terminal then concatenates the sleep quality parameters and the activity trend parameters to obtain the contextual feature vector of the sleep pattern.

[0095] S305, based on the instantaneous context feature vector, the activity cycle context feature vector, and the work-rest pattern context feature vector, a comprehensive context feature vector is obtained.

[0096] Specifically, the terminal concatenates the instantaneous context feature vector, the activity cycle context feature vector, and the daily routine context feature vector to obtain a comprehensive context feature vector.

[0097] The health information management method based on exercise data provided in this embodiment extracts and fuses features from real-time exercise physiological data, real-time activity status tags, and daily routine data at multiple time scales to obtain a comprehensive contextual feature vector that fully describes the user's physiological state. This method considers both immediate physiological reactions and personal activity habits and long-term health trends, providing rich input information for accurate health warnings and improving the dimensionality and accuracy of health status assessment.

[0098] In one embodiment, the daily routine data includes at least the sleep duration parameter, the deep sleep ratio parameter, the number of times the person wakes up, the average number of steps on the first day, and the average number of steps on the second day; the sampling timestamp of the average number of steps on the first day is later than the sampling timestamp of the average number of steps on the second day.

[0099] Based on daily routine data, we analyze the changing trends of users' daily routines to obtain contextual feature vectors of daily routines, including:

[0100] S401, based on a preset sleep parameter database, normalizes daily routine data to obtain normalized daily routine data; the normalized daily routine data includes normalized sleep duration parameters, normalized deep sleep ratio parameters, and normalized number of awakenings.

[0101] Specifically, the preset sleep parameter database includes the mean and standard deviation of sleep duration parameters, the mean and standard deviation of deep sleep ratio parameters, and the mean and standard deviation of wake-up frequency parameters for a large population. This database can be collected based on actual work experience, and this embodiment does not further limit the setting of this preset sleep parameter database. The terminal uses the z-score method to standardize the sleep duration parameters, deep sleep ratio parameters, and wake-up frequency parameters in the current user's daily routine data based on the mean and standard deviation of sleep duration parameters, deep sleep ratio parameters, and wake-up frequency parameters from the preset sleep parameter database. This results in normalized sleep duration parameters, normalized deep sleep ratio parameters, and normalized wake-up frequency parameters. These parameters are then concatenated to obtain normalized daily routine data.

[0102] S402, based on normalized daily routine data, uses the following formula to calculate sleep quality parameters:

[0103]

[0104] in, It is a sleep quality parameter. It's a time-based weighting. It is a normalized sleep duration parameter. It is deep sleep weight. It is the normalized deep sleep ratio parameter. It's the weight of waking up. It is the normalized number of times you wake up.

[0105] Specifically, the terminal uses a formula to calculate sleep quality parameters based on normalized sleep duration, normalized deep sleep percentage, and normalized wakefulness frequency from normalized daily routine data. Among these parameters... Used to characterize the current user's sleep quality. A value closer to 1 indicates better sleep quality, while a value closer to 0 indicates poorer sleep quality. Normalized sleep duration parameter. Normalized deep sleep ratio parameter and normalized wake-up times This can be obtained from normalized daily routine data derived from S401. Duration weighting. Deep sleep weights are used to characterize the impact of normalized sleep duration parameters on sleep quality parameters. Wake-up weight is used to characterize the impact of normalized sleep duration parameters on sleep quality parameters. Used to characterize the impact on sleep quality parameters, the sum of duration weight, deep sleep weight, and wakefulness weight is 1, and can be set according to actual work.

[0106] S403, based on the average steps of the first day and the average steps of the second day, uses the following formula to calculate the activity trend parameter:

[0107]

[0108] in, It is an activity level trend parameter. It is the average number of steps taken on the first day. This is the average number of steps taken on the second day.

[0109] Specifically, the terminal calculates the activity trend parameter based on the average steps taken on the first day and the average steps taken on the second day, according to a formula. The sampling timestamp for the average steps taken on the first day is later than that for the average steps taken on the second day, indicating that the average steps taken on the first day represent the activity state closer to the current sampling timestamp. Activity Trend Parameter It indicates the direction and magnitude of the current user activity level change. This indicates an increase in activity level. This indicates a decrease in activity level. Average steps taken on the first day. Average steps per day and day This can be obtained based on daily routine data.

[0110] S404, based on sleep quality parameters and activity trend parameters, obtains the contextual feature vector of daily routine.

[0111] Specifically, the terminal concatenates sleep quality parameters and activity trend parameters to obtain a contextual feature vector of sleep patterns.

[0112] The health information management method based on exercise data provided in this embodiment obtains a quantified contextual feature vector of daily routines through sleep quality assessment and activity trend analysis, providing a more comprehensive perspective on health status. This allows health warnings to not only be based on the current physiological state, but also to consider personal lifestyle habits and long-term trends, thereby improving the depth and foresight of health risk assessment.

[0113] In one embodiment, the causal relationship map is obtained through the following method:

[0114] S501, retrieve the preset historical physiological database; the historical physiological database includes the user's historical sleep quality parameters, average daily steps, and heart rate parameters with real-time activity status tags.

[0115] Specifically, the pre-set historical physiological database includes the user's historical sleep quality parameters, average daily steps, and heart rate parameters with real-time activity status tags. The pre-set historical physiological database is organized in time series format, with each data point containing a sampling timestamp and the corresponding parameter value.

[0116] S502 uses heart rate parameters, sleep quality parameters, and average daily steps as nodes to construct an initial causal relationship graph.

[0117] Specifically, the terminal uses heart rate parameters, sleep quality parameters, and average daily steps as nodes to construct an initial causal relationship graph based on a pre-set medical knowledge base. The mathematical representation of this initial causal relationship graph is as follows: ,in It is a set of nodes. The real-time activity status label is Heart rate parameters, These are various sleep quality parameters. This is the average number of steps taken each day; This refers to the initial connection relationships between nodes. For example, a preset medical knowledge base is used to characterize the influence relationship between any two graph nodes in the node set, such as sleep quality affecting heart rate and activity level affecting sleep quality, which can be set based on medical knowledge.

[0118] S503, based on the initial causal relationship graph, performs independence verification on each node in the initial causal relationship graph to obtain the correlation coefficient between each node; and constructs the connection edges between each node based on the correlation coefficient to obtain the complete causal relationship graph.

[0119] Specifically, the terminal uses statistical methods to verify the independence between nodes based on the initial causal relationship graph. This is achieved by calculating partial correlation coefficients. To test the control of other variables Under the condition, variables and Whether they are independent. And based on the correlation coefficient, connecting edges are added between nodes in the initial graph to obtain a complete causal relationship graph. Among them, , and It can be any graph node type. For example, suppose that at this point, (Sleep quality) (Heart rate). Condition set It can be Except Any subset outside, in this example, One possible option is (Average daily steps). The test requires calculating the partial correlation coefficient. It measures the degree of control (or "fixation"). After the impact, and The remaining correlation between them.

[0120] S504. For each connecting edge in the complete causal relationship graph, analyze the causal relationship between the nodes at both ends of the connecting edge to obtain the causal relationship parameters. The causal relationship parameters include the causal relationship pointing value and the relationship strength. The causal relationship pointing value is used to indicate the causal relationship between nodes.

[0121] Specifically, the terminal employs a dual machine learning approach to estimate causal effects for each directed connection edge in the complete causal graph, thereby quantifying the causal relationships between nodes and obtaining causal relationship parameters including causal relationship direction values ​​and relationship strengths. Dual machine learning is an econometric method for estimating causal effects, effectively controlling for confounding variables (i.e., factors that may simultaneously influence the two variables besides the two variables being analyzed) through a two-stage machine learning model. Specifically, for each edge in the graph... The terminal randomly divides the historical physiological database into two independent subsets. and Use first Train two gradient boosting tree models, namely model 1. and model :Model Used for fitting variables For all other covariates (Right now The conditional expectation of ), i.e. ;Model Used to fit the result variable For all other covariates The conditional expectation, i.e. Reuse Calculate residuals: Processing variable residuals and outcome variable residuals And through linear regression To estimate the causal effect coefficient This coefficient This constitutes the causal relationship parameters of the connecting edges, where the sign (positive or negative) of the causal relationship parameter is converted into a causal relationship pointing value, and the relationship strength indicates the direction of the causal action (e.g., if...). A negative value indicates that decreased sleep quality leads to an increased heart rate; the absolute value of the causal relationship parameter is determined as the relationship strength, which quantifies the degree to which each unit change in the causal variable leads to a change in the outcome variable (e.g., This indicates that for every 1 unit decrease in the sleep quality index, the expected heart rate increases by 3.2 bpm.

[0122] S505, adjust and improve the causal relationship map based on the causal relationship parameters to obtain the causal relationship map.

[0123] Specifically, the terminal sets the connection edges and edge weights between corresponding graph nodes in the complete causal relationship graph based on the causal relationship pointing value and relationship strength included in the causal relationship parameters, thereby obtaining the causal relationship graph.

[0124] This embodiment provides a health information management method based on exercise data. By combining statistical correlation analysis with causal inference techniques (dual machine learning), it automatically mines and quantifies stable causal relationships between key health indicators from historical data, obtaining a causal relationship graph. In this graph, nodes represent core health indicators, connecting edges represent directed causal influences, and the weights on the edges precisely quantify the intensity of the influence. It provides a standard for predictively adjusting downstream target variables based on changes in the state of upstream causal variables, offering a core knowledge foundation for achieving personalized, interpretable, and dynamic early warning.

[0125] In one embodiment, a dynamic warning threshold is generated based on real-time motion physiological data, a comprehensive contextual feature vector, and a preset causal relationship graph; and interpretable warning information is generated based on the dynamic warning threshold and real-time motion physiological data, including:

[0126] S601 calculates the basic warning threshold based on the activity cycle context feature vector, the instantaneous context feature vector, and the preset scaling factor.

[0127] Specifically, the terminal uses the following formula based on the current user's activity cycle context feature vector, instantaneous context feature vector, and preset scaling factor: The basic early warning threshold is calculated. This is the basic early warning threshold. It is a real-time data weight used to characterize the impact of instantaneous context feature vectors on the basic early warning threshold. This is a preset scaling factor used to characterize the relationship between the baseline threshold and the fluctuation range of heart rate parameters. The default setting is 2.5, but it can also be set according to actual work. In this embodiment, the default setting is 2.5, and the setting is not further limited. It is the mean of the heart rate parameter in the instantaneous context feature vector. It is the variance of the heart rate parameter in the instantaneous context feature vector. It is the mean of the heart rate parameter in the current cycle within the activity cycle context feature vector. It is the variance of the heart rate parameter in the current cycle within the activity cycle context feature vector.

[0128] S602, based on the causal relationship graph, extracts the nodes and causal relationship parameters that affect heart rate parameters, and obtains a real-time quantitative set of influence relationships.

[0129] Specifically, the terminal queries a preset causal relationship graph to identify all variables in the graph that are associated with the target warning variable—heart rate parameter. — A directed edge that serves as the endpoint. For each such edge Extract its starting node (i.e., variables that affect heart rate, such as sleep quality parameters) ) and the causal relationship parameters corresponding to that edge. This forms a real-time quantitative set of influence relationships.

[0130] S603, based on the real-time quantitative influence relationship set and the preset historical parameter mean, quantifies the influence of the context feature vector of work and rest patterns on the basic early warning threshold, and obtains the total compensation value of causal effect.

[0131] Specifically, the terminal utilizes real-time quantification of information in the influence relation set to calculate the expected total impact on heart rate due to upstream causal variables deviating from their baseline, thus obtaining the total compensation value for the causal effect. The terminal is based on the additive assumption of causal effects (which holds under the linear model approximation of the causal relationship map). For each item in the relation set... Based on the parameters representing the current state included in the contextual feature vector of the daily routine, calculate the expected change in heart rate caused by this causal variable: And sum the expected effects of all causal variables: The total causal effect compensation value is obtained. The unit of the total causal effect compensation value is bpm, and its physical meaning is: the "reasonable" deviation of the heart rate relative to its baseline level, which is purely determined by the type of activity, based on the user's current activity pattern (such as sleep quality and activity trend).

[0132] S604, based on the total compensation value of causal effect and the basic early warning threshold, the dynamic early warning threshold is calculated using the following formula:

[0133]

[0134] in, It is a dynamic early warning threshold. This is the basic early warning threshold. It is an adjustment factor. It is the total compensation value of causal effect.

[0135] Specifically, the terminal uses a formula to weight and combine the total causal effect compensation value and the basic early warning threshold to obtain the dynamic early warning threshold. The dynamic early warning threshold... It is a heart rate parameter threshold that integrates the user's real-time activity type, short-term physiological fluctuations, and long-term state effects explained by causality. Baseline warning threshold. This can be obtained from S601. Adjustment coefficient. This is an adjustable parameter between 0 and 1, used to control the strength of incorporating the expected offset of the causal explanation into the threshold adjustment, and can be set according to the actual work. Total compensation value for causal effects. It can be obtained from S603.

[0136] S605 compares the dynamic warning threshold with real-time motion physiological data to obtain the physiological parameter comparison result; when the physiological parameter comparison result meets the preset warning requirements, a warning sign is obtained.

[0137] Specifically, the terminal compares the heart rate parameter in the current user's real-time exercise physiological data with a dynamic warning threshold. When the comparison result meets the preset warning requirements, a warning indicator is generated. The preset warning requirement is that the heart rate parameter in the real-time exercise physiological data is greater than the dynamic warning threshold.

[0138] S606 analyzes the contribution of the contextual feature vector of the work-rest pattern to the total compensation value of the causal effect based on the early warning identifier, real-time quantitative impact relationship set and causal effect total compensation value, and obtains the contribution ratio parameter; and generates interpretable early warning information based on the dynamic early warning threshold, physiological parameter comparison results and contribution ratio parameter.

[0139] Specifically, the terminal analysis of the total compensation value of causal effects The main source is the set of real-time quantitative impact relationships, from which the expected impact amount is identified. The causal variable with the largest absolute value Calculate its contribution ratio The contribution ratio parameter is obtained. Then, the dynamic warning threshold, physiological parameter comparison results, and contribution ratio parameter are assembled into a natural language statement according to preset language organization rules to obtain interpretable warning information. For example, the preset language organization rules can be set according to the format of warning information in actual work; the form of the preset language organization rules could be "Warning: Your heart rate ([ [bpm] has exceeded the limit based on your recent [ The adjusted safety threshold ([A] bpm) is abnormal. This threshold increase ([ bpm) mainly ([ ]%) attributed to '[ The impact of ']. Among them It is a dynamic early warning threshold.

[0140] The health information management method based on exercise data provided in this embodiment integrates a multi-granularity temporal context model with a causal relationship graph. Through a causal effect compensation mechanism, it dynamically and personally adjusts traditional activity baseline thresholds to obtain dynamic warning thresholds. Furthermore, it generates interpretable warning information based on these dynamic warning thresholds. This achieves dynamic threshold adjustment based on individual physiological characteristics and activity context, overcoming the high false alarm rate of traditional static threshold methods under different activity states. It improves the accuracy and personalization of health risk warnings, while the interpretable warning information enhances user trust and compliance.

[0141] The aforementioned health information management method based on exercise data acquires the user's real-time exercise physiological data and daily routine data. Real-time exercise physiological data includes heart rate parameters, accelerometer parameters, and gyroscope parameters. Daily routine data characterizes the user's daily routine. Based on the real-time exercise physiological data, the user's activity state is analyzed to obtain a real-time activity state label. This label characterizes the user's real-time activity type. Based on the real-time exercise physiological data, real-time activity state labels, and daily routine data, the characteristics of the user's physical changes are analyzed to obtain a comprehensive contextual feature vector. This vector includes instantaneous contextual feature vectors, activity cycle contextual feature vectors, and daily routine pattern contextual feature vectors. Based on the real-time exercise physiological data, the comprehensive contextual feature vector, and a preset causal relationship graph, a dynamic warning threshold is generated. And based on the dynamic warning threshold and real-time exercise physiological data, interpretable warning information is generated. This method achieves dynamic threshold adjustment based on individual physiological characteristics and activity context, overcoming the high false alarm rate of traditional static threshold methods under different activity states. It improves the accuracy and personalization of health risk warnings, while enhancing user trust and compliance through interpretable warning information.

[0142] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0143] Based on the same inventive concept, this application also provides a health information management system based on exercise data for implementing the above-mentioned health information management method based on exercise data. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the health information management system based on exercise data provided below can be found in the limitations of the health information management method based on exercise data described above, and will not be repeated here.

[0144] In one exemplary embodiment, such as Figure 2 As shown, a health information management system 200 based on exercise data is provided, including:

[0145] Data acquisition module 201 is used to acquire the user's real-time exercise physiological data and daily routine data; the real-time exercise physiological data includes heart rate parameters, accelerometer parameters, and gyroscope parameters; the daily routine data is used to characterize the user's daily routine.

[0146] The status analysis module 202 is used to analyze the user's activity status based on real-time motion physiological data and obtain real-time activity status labels; the real-time activity status labels are used to characterize the user's real-time activity type.

[0147] The state feature extraction module 203 is used to analyze the characteristics of user's physical changes based on real-time exercise physiological data, real-time activity status labels and daily routine data, and obtain a comprehensive context feature vector; the comprehensive context feature vector includes instantaneous context feature vector, activity cycle context feature vector and daily routine context feature vector.

[0148] The early warning information generation module 204 is used to generate dynamic early warning thresholds based on real-time exercise physiological data, comprehensive contextual feature vectors, and preset causal relationship maps; and to generate interpretable early warning information based on dynamic early warning thresholds and real-time exercise physiological data; the interpretable early warning information is used to characterize the health risks and causes of the user.

[0149] Furthermore, the state analysis module is also used for:

[0150] Based on a preset physiological data threshold, real-time exercise physiological data is filtered to obtain filtered exercise physiological data.

[0151] Based on a preset detection time window, the filtered motion physiological data is segmented to obtain the motion physiological data to be tested; the motion physiological data to be tested includes the heart rate parameters, accelerometer parameters, and gyroscope parameters to be tested.

[0152] Based on the accelerometer parameters to be measured, the user's activity intensity is quantified to obtain activity intensity parameters;

[0153] Calculate the variance of the parameters of the gyroscope under test to obtain the gyroscope variance parameters; calculate the variance of the parameters of the accelerometer under test to obtain the acceleration variance parameters.

[0154] The exercise physiological data to be tested is transformed in the frequency domain to obtain the exercise physiological spectrum; and the spectral entropy of the exercise physiological spectrum is calculated to obtain the data complexity parameter.

[0155] The physiological spectrum of the motion to be measured, the gyroscope variance parameter, the acceleration variance parameter, and the data complexity parameter are input into the preset activity state classification model to obtain the real-time activity state label; the activity state classification model is constructed based on the random forest classifier.

[0156] Furthermore, the state feature extraction module includes:

[0157] The instantaneous feature extraction unit is used to calculate the mean and variance of heart rate parameters based on a preset instantaneous time window and a preset instantaneous time decay factor, and obtain the instantaneous context feature vector.

[0158] The periodic reference selection unit is used to select instantaneous context feature vectors from preset historical motion physiological data based on real-time activity status labels and preset periodic time windows to form a periodic instantaneous context feature vector set; the historical motion physiological data includes the real-time activity status labels of the user's historical motion and the corresponding instantaneous context feature vectors;

[0159] The periodic feature extraction unit is used to calculate the mean and variance of the instantaneous context feature vectors in the periodic instantaneous context feature vector set based on the preset periodic time decay factor, so as to obtain the activity periodic context feature vector.

[0160] The daily routine feature extraction unit is used to analyze the changing trends of users' daily routines based on daily routine data, and obtain the contextual feature vector of daily routines;

[0161] The comprehensive feature generation unit is used to obtain a comprehensive context feature vector based on the instantaneous context feature vector, the activity cycle context feature vector, and the work-rest pattern context feature vector.

[0162] Furthermore, the daily routine data includes at least the parameters of sleep duration, deep sleep ratio, number of times the person wakes up, average steps on the first day and average steps on the second day; the sampling timestamp of the average steps on the first day is later than the sampling timestamp of the average steps on the second day.

[0163] The daily routine feature extraction unit is also used for:

[0164] Based on a pre-set sleep parameter database, the daily routine data is normalized to obtain normalized daily routine data; the normalized daily routine data includes normalized sleep duration parameters, normalized deep sleep ratio parameters, and normalized number of awakenings.

[0165] Based on normalized daily routine data, sleep quality parameters are calculated using the following formula:

[0166]

[0167] in, It is a sleep quality parameter. It's a time-based weighting. It is a normalized sleep duration parameter. It is deep sleep weight. It is the normalized deep sleep ratio parameter. It's the weight of waking up. It is the normalized number of times you wake up;

[0168] Based on the average steps taken on the first day and the average steps taken on the second day, the activity trend parameter is calculated using the following formula:

[0169]

[0170] in, It is an activity level trend parameter. It is the average number of steps taken on the first day. This is the average number of steps taken on the second day;

[0171] Based on sleep quality parameters and activity trend parameters, a contextual feature vector of daily routine patterns is obtained.

[0172] Furthermore, the system includes a map generation module for:

[0173] Obtain a preset historical physiological database; the historical physiological database includes the user's historical sleep quality parameters, average daily steps, and heart rate parameters with real-time activity status tags;

[0174] Using heart rate parameters, sleep quality parameters, and average daily steps as nodes, an initial causal relationship graph was constructed.

[0175] Based on the initial causal relationship graph, the independence of each node in the initial causal relationship graph is verified to obtain the correlation coefficient between each node; and based on the correlation coefficient, the connection edges between each node are constructed to obtain the complete causal relationship graph.

[0176] For each connecting edge in the complete causal relationship graph, the causal relationship between the nodes at both ends of the connecting edge is analyzed to obtain the causal relationship parameters. The causal relationship parameters include the causal relationship pointing value and the relationship strength. The causal relationship pointing value is used to indicate the causal relationship between nodes.

[0177] The causal relationship map is obtained by adjusting and improving the causal relationship parameters.

[0178] Furthermore, the early warning information generation module is also used for:

[0179] The basic warning threshold is calculated based on the activity cycle context feature vector, the instantaneous context feature vector, and the preset scaling factor.

[0180] Based on the causal relationship graph, nodes and causal relationship parameters affecting heart rate parameters are extracted to obtain a real-time quantitative set of influence relationships;

[0181] Based on the real-time quantitative impact relationship set and the preset historical parameter mean, the impact of the context feature vector of work and rest pattern on the basic early warning threshold is quantified to obtain the total compensation value of causal effect;

[0182] Based on the total compensation value of causal effects and the basic early warning threshold, the dynamic early warning threshold is calculated using the following formula:

[0183]

[0184] in, It is a dynamic early warning threshold. This is the basic early warning threshold. It is an adjustment factor. It is the total compensation value of causal effects;

[0185] The dynamic warning threshold and real-time motion physiological data are compared to obtain the physiological parameter comparison results; when the physiological parameter comparison results meet the preset warning requirements, a warning sign is obtained.

[0186] Based on early warning indicators, real-time quantitative impact relationship sets, and total causal effect compensation values, the contribution of the contextual feature vector of daily routines to the total causal effect compensation value is analyzed to obtain contribution ratio parameters. Based on dynamic early warning thresholds, physiological parameter comparison results, and contribution ratio parameters, interpretable early warning information is generated.

[0187] In one embodiment, such as Figure 3 A computer device is provided, comprising:

[0188] At least one processor 301, and a memory 302 communicatively connected to at least one of the processors 301: the memory stores application code executable by at least one of the processors, the application code being executed by at least one of the processors to enable at least one of the processors to perform the health information management method based on motion data as described above.

[0189] Computer equipment may also include: sensor 303.

[0190] The processor 301, memory 302 and sensor 303 can be connected via a bus or other means, with the bus being an example in the figure.

[0191] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0192] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and 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 modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0193] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A health information management method based on motion data, characterized by, The method comprises: acquiring real-time motion physiological data and life schedule data of a user; the real-time motion physiological data comprises heart rate parameters, accelerometer parameters and gyroscope parameters; the life schedule data is used to represent the schedule of the user; based on the real-time motion physiological data, analyzing the activity state of the user to obtain a real-time activity state label; the real-time activity state label is used to represent the real-time activity type of the user; based on the real-time motion physiological data, the real-time activity state label and the life schedule data, analyzing the features of the body changes of the user to obtain a comprehensive context feature vector; the comprehensive context feature vector comprises an instantaneous context feature vector, an activity period context feature vector and a schedule rule context feature vector; based on the real-time motion physiological data, the comprehensive context feature vector and a preset causal relationship graph, generating a dynamic early warning threshold; and based on the dynamic early warning threshold and the real-time motion physiological data, generating interpretable early warning information; the interpretable early warning information is used to represent the health risks and reasons existing in the user.

2. The method of claim 1, wherein, The method comprises: based on a preset physiological data threshold, filtering the real-time motion physiological data to obtain filtered motion physiological data; based on a preset detection time window, segmenting the filtered motion physiological data to obtain to-be-detected motion physiological data; the to-be-detected motion physiological data comprises to-be-detected heart rate parameters, to-be-detected accelerometer parameters and to-be-detected gyroscope parameters; based on the to-be-detected accelerometer parameters, quantifying the activity intensity of the user to obtain an activity intensity parameter; calculating the variance of the to-be-detected gyroscope parameters to obtain a gyroscope variance parameter; and calculating the variance of the to-be-detected accelerometer parameters to obtain an acceleration variance parameter; performing frequency domain conversion on the to-be-detected motion physiological data to obtain to-be-detected motion physiological frequency spectrum; and calculating the spectral entropy of the to-be-detected motion physiological frequency spectrum to obtain a data complexity parameter; inputting the to-be-detected motion physiological frequency spectrum, the gyroscope variance parameter, the acceleration variance parameter and the data complexity parameter into a preset activity state classification model to obtain the real-time activity state label; the activity state classification model is constructed based on a random forest classifier.

3. The method of claim 1, wherein, The method comprises: based on a preset instantaneous time window and a preset instantaneous time decay factor, calculating the mean and variance of the heart rate parameters to obtain the instantaneous context feature vector; based on the real-time activity state label and a preset period time window, selecting the instantaneous context feature vector from preset historical motion physiological data to form a period instantaneous context feature vector set; the historical motion physiological data comprises the real-time activity state label and the corresponding instantaneous context feature vector of the historical motion of the user; calculating a mean and a variance of the instantaneous context feature vectors in the periodic instantaneous context feature vector set based on a preset periodic time decay factor, to obtain the activity periodic context feature vector; analyzing a change trend of a living and resting rule of the user based on the living and resting data, to obtain the living and resting rule context feature vector; obtaining the comprehensive context feature vector based on the instantaneous context feature vector, the activity periodic context feature vector and the living and resting rule context feature vector.

4. The method of claim 3, wherein, The living and resting data at least includes a sleep duration parameter, a deep sleep proportion parameter, a wake-up times, a first daily average step number and a second daily average step number; a sampling timestamp of the first daily average step number is later than a sampling timestamp of the second daily average step number; The analyzing a change trend of a living and resting rule of the user based on the living and resting data, to obtain the living and resting rule context feature vector, includes: normalizing the living and resting data based on a preset sleep parameter database, to obtain normalized living and resting data; the normalized living and resting data includes a normalized sleep duration parameter, a normalized deep sleep proportion parameter and a normalized wake-up times; calculating a sleep quality parameter based on the normalized living and resting data using the following formula: wherein, is a sleep quality parameter, is a duration weight, is a normalized sleep duration parameter, is a deep sleep weight, is a normalized deep sleep proportion parameter, is a wake-up weight, is a normalized wake-up count; calculating an activity amount trend parameter based on the first daily average step number and the second daily average step number using the following formula: wherein, is an activity amount trend parameter, is a first daily average step count, is a second daily average step count; obtaining the living and resting rule context feature vector based on the sleep quality parameter and the activity amount trend parameter.

5. The method of claim 4, wherein, The causal relationship graph is obtained by the following method: obtaining a preset historical physiological database; the historical physiological database includes the sleep quality parameter, the daily average step number and the heart rate parameter with the real-time activity state label of the user history; constructing an initial causal relationship graph with the heart rate parameter, the sleep quality parameter and the daily average step number as nodes; verifying the independence of each node in the initial causal relationship graph based on the initial causal relationship graph, to obtain a correlation coefficient between each node; and constructing a connection edge between each node based on the correlation coefficient, to obtain a perfect causal relationship graph; analyzing the causal relationship of the nodes at both ends of each connection edge in the perfect causal relationship graph, to obtain a causal relationship parameter; the causal relationship parameter includes a causal relationship pointing value and a relationship strength; the causal relationship pointing value is used to indicate the causal relationship between the nodes; adjusting the perfect causal relationship graph based on the causal relationship parameter, to obtain the causal relationship graph.

6. The method of claim 5, wherein, generating a dynamic early warning threshold based on the real-time exercise physiological data, the comprehensive context feature vector and a preset causal relationship graph; and generating interpretable early warning information based on the dynamic early warning threshold and the real-time exercise physiological data, including: calculating a basic early warning threshold based on the activity periodic context feature vector, the instantaneous context feature vector and a preset scaling factor; extracting the nodes affecting the heart rate parameter and the causal relationship parameter based on the causal relationship graph, to obtain a real-time quantitative influence relationship set; quantify an influence of the sleep-wake regularity context feature vector on the base early warning threshold based on the real-time quantified influence relationship set and a preset historical parameter mean value, to obtain a total causal effect compensation value; based on the total causal effect compensation value and the base early warning threshold, calculate the dynamic early warning threshold using the following formula: wherein, is a dynamic early warning threshold, is a base early warning threshold, is an adjustment coefficient, is a total causal effect compensation value; compare the dynamic early warning threshold and the real-time exercise physiological data to obtain a physiological parameter comparison result; when the physiological parameter comparison result meets a preset warning requirement, obtain an early warning identifier; based on the early warning identifier, the real-time quantified influence relationship set and the total causal effect compensation value, analyze a contribution of the sleep-wake regularity context feature vector to the total causal effect compensation value, to obtain a contribution proportion parameter; and based on the dynamic early warning threshold, the physiological parameter comparison result and the contribution proportion parameter, generate the interpretable early warning information.

7. A health information management system based on motion data, characterized by, The system comprises: a data acquisition module configured to acquire real-time exercise physiological data and life and work data of a user; the real-time exercise physiological data comprises heart rate parameters, accelerometer parameters and gyroscope parameters; and the life and work data is used to represent the work and rest situation of the user; a state analysis module configured to analyze an activity state of the user based on the real-time exercise physiological data, to obtain a real-time activity state label; the real-time activity state label is used to represent a real-time activity type of the user; a state feature extraction module configured to analyze features of body changes of the user based on the real-time exercise physiological data, the real-time activity state label and the life and work data, to obtain a comprehensive context feature vector; the comprehensive context feature vector comprises an instantaneous context feature vector, an activity period context feature vector and a sleep-wake regularity context feature vector; an early warning information generation module configured to generate a dynamic early warning threshold based on the real-time exercise physiological data, the comprehensive context feature vector and a preset causal relationship graph; and generate interpretable early warning information based on the dynamic early warning threshold and the real-time exercise physiological data; the interpretable early warning information is used to represent a health risk and a reason existing in the user.

8. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that, The processor executes the computer program to implement the steps of the method of any one of claims 1 to 6.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 6.

Images