Health monitoring method and system of intelligent wearable device
By preprocessing and residual reduction of health monitoring data from smart wearable devices, the problem of abnormal data affecting GBDT modeling is solved, and more accurate sleep state assessment and reminder functions are achieved.
Patent Information
- Application Number
- CN202511871550.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-12
- Publication Date
- 2026-01-09
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In existing smart wearable devices, abnormal data in health monitoring leads to inaccurate GBDT modeling, affecting the accuracy of sleep state assessment.
By collecting and preprocessing health monitoring data, calculating the residual sequence and trend of each data point, obtaining the probability of anomalies, and using the residual reduction index to weaken the residuals in GBDT modeling, the impact of outlier data is reduced, and a more accurate GBDT model is established.
This enables more accurate monitoring of sleep health status of smart wearable devices, reduces the impact of abnormal data on modeling, and improves the accuracy of the model and the reliability of sleep assessment.
Smart Images
Figure CN121306480A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of data processing technology, specifically to a health monitoring method and system for a smart wearable device. Background Technology
[0002] With the increasing global aging population, the high incidence of chronic diseases, and the pressures of a fast-paced lifestyle, health management needs are shifting from "disease treatment" to "full-cycle health intervention." Smart wearable devices, with their non-intrusive monitoring and real-time data feedback capabilities, are becoming a core driving force for innovation in the health industry. In the future, with breakthroughs in implantable sensors and non-invasive blood glucose monitoring technology, the monitoring of health, especially sleep health, will require even higher standards and requirements.
[0003] Currently, health monitoring data from smart wearable devices may contain anomalies, making it difficult to achieve accurate initial GBDT modeling. In the process of health monitoring of smart wearable devices using the conventional GBDT algorithm, the modeling process for sleep state assessment through analysis of various monitoring data has high requirements for data quality. Anomalies in the health monitoring devices can lead to large residuals in the calculations, causing errors in the model's fit to these anomalies and resulting in inaccurate detection of subsequent data. Summary of the Invention
[0004] To reduce the impact of outlier data on the GBDT algorithm and achieve more accurate GBDT modeling, this application provides a health monitoring method and system for smart wearable devices.
[0005] Firstly, this application provides a health monitoring method for a smart wearable device, employing the following technical solution:
[0006] Collect health monitoring data from smart wearable devices and preprocess the health monitoring data;
[0007] The preprocessed health monitoring data is iterated according to the GBDT algorithm to obtain the residual sequence of each data point corresponding to the preprocessed health monitoring data. Here, all health monitoring data items collected in one session constitute a data point.
[0008] Based on the residual sequence, the residual trend of each data point corresponding to the preprocessed health monitoring data is obtained;
[0009] Based on the distribution of the preprocessed health monitoring data, the probability of anomalies for each data point corresponding to the preprocessed health monitoring data is obtained.
[0010] Based on the probability of anomalies and the residual trend of each data point corresponding to the preprocessed health monitoring data, the corresponding residual reduction index is obtained;
[0011] The residuals of the preprocessed health monitoring data during GBDT modeling are weakened according to the residual reduction index.
[0012] The sleep health status of smart wearable devices is monitored based on the GBDT model with residual weakening.
[0013] The process of collecting health monitoring data from smart wearable devices and preprocessing the health monitoring data includes:
[0014] Collect sleep monitoring data from each session as health monitoring data;
[0015] The first The first collection The data item is denoted as The total number of sleep data collections is denoted as n, and the number of data items collected in each sleep monitoring session is denoted as n. ;
[0016] For each data point, a sleep score is calculated based on all health monitoring data. The score will be determined by the data point numbered... The sleep score corresponding to the data points collected each time is recorded as follows: GBDT modeling was performed using historical health monitoring data to predict subsequent sleep scores.
[0017] The step of iterating through the preprocessed health monitoring data according to the GBDT algorithm to obtain the residual sequence of each preprocessed health monitoring data includes:
[0018] The GBDT algorithm is used to perform initial modeling iterations on the sample health monitoring data, where the preset initial iteration number is [number missing]. ;
[0019] The first The data point at the th th The absolute value of the residual after the nth iteration is denoted as , among which, the The residual data of all data points are sorted according to the iteration number to form the... The residual sequence of data points.
[0020] The calculation formula corresponding to the probability of an anomaly includes:
[0021]
[0022] in, Indicates the first The probability of anomalies in each data point Indicates the first The number of temporal and spatial neighborhood data points for each data point Indicates the first The first data point The Euclidean distance between a time-series neighborhood data point and its Euclidean distance. Indicates the first The data point and its first Euclidean distance between spatial neighborhood data points Indicates the first The average Euclidean distance between time-series neighboring data points of each data point Indicates the first The average Euclidean distance between spatial neighbor data points of a given data point.
[0023] The calculation formula corresponding to the residual trend includes:
[0024]
[0025] in, Indicates the first Residual trend of each data point and They represent the first The residual sequence of the nth data point The and the first One data point, Indicates the first The range of the residual sequence of data points. This represents the ratio of the difference between two adjacent residual data points to the average difference. This represents the difference in residual data between two consecutive iterations.
[0026] The process of reducing the residuals of the preprocessed health monitoring data during GBDT modeling based on the residual reduction index includes:
[0027] The ratio of the residual trend to the probability of anomalies of a data point is used as the residual reduction index for that data point.
[0028] All data points whose residual reduction index is less than the preset residual reduction index threshold are recorded as data points to be reduced;
[0029] The residual obtained during the GBDT algorithm iteration process of the data points to be reduced is multiplied by the residual reduction exponent to obtain the reduced residual;
[0030] Based on the reduced residuals, GBDT iteration is performed to obtain the corresponding prediction model.
[0031] The monitoring of sleep health status of smart wearable devices based on the residual-weakened GBDT model includes:
[0032] Predict sleep scores for smart wearable devices based on the GBDT model with residual weakening;
[0033] If the sleep score is lower than the preset sleep score threshold, it is determined that the human sleep is abnormal and a timely reminder is issued.
[0034] Secondly, this application provides a health monitoring system for a smart wearable device, which adopts the following technical solution:
[0035] The data acquisition module is used to collect health monitoring data from smart wearable devices and to preprocess the health monitoring data.
[0036] The residual sequence calculation module is used to iterate the preprocessed health monitoring data according to the GBDT algorithm to obtain the residual sequence of each data point corresponding to the preprocessed health monitoring data, wherein all health monitoring data items collected in one time constitute a data point;
[0037] The residual trend calculation module is used to obtain the residual trend of each data point corresponding to the preprocessed health monitoring data based on the residual sequence.
[0038] An anomaly probability calculation module is used to obtain the anomaly probability of each data point corresponding to the preprocessed health monitoring data based on the distribution of the preprocessed health monitoring data.
[0039] The residual reduction index calculation module obtains the corresponding residual reduction index based on the probability of anomalies and the residual trend of each data point corresponding to the preprocessed health monitoring data.
[0040] The residual reduction module is used to reduce the residuals of the preprocessed health monitoring data during the GBDT modeling process according to the residual reduction index.
[0041] The monitoring module is used to monitor the sleep health status of smart wearable devices based on the GBDT model after residual reduction.
[0042] This application has the following technical effects: In the process of monitoring sleep health status using health monitoring data collected by wearable devices using the conventional GBDT algorithm, GBDT modeling may be affected by abnormal data, leading to inaccurate modeling. This application calculates the probability of abnormality for each data point in the sample data, and then performs dynamic reduction of the residuals based on the changes in the residuals during the GBDT modeling process. This reduces the impact of abnormal data on the GBDT algorithm and achieves more accurate GBDT modeling. Attached Figure Description
[0043] Figure 1 This is a flowchart illustrating steps S1-S7 of a health monitoring method for a smart wearable device according to this application. Detailed Implementation
[0044] This application discloses a health monitoring method for a smart wearable device, referring to... Figure 1 ,include:
[0045] S1. Collect health monitoring data from smart wearable devices and preprocess the health monitoring data.
[0046] In one embodiment, health monitoring data from a smart wearable device is collected, including data related to sleep health. Specifically, the acquired data includes sleep duration, number of awakenings, highest / lowest heart rate, blood oxygen saturation, body temperature, and respiratory rate, which can be used to monitor sleep status. Each sleep monitoring session is collected as health monitoring data. The first collection The data item is denoted as A single data point is formed by collecting all data items in one session. Let n be the total number of sleep data collections, and let n be the number of data items collected in each session. .
[0047] Furthermore, a sleep score is calculated for all health monitoring data corresponding to each data point, and the first... The sleep score corresponding to the data points collected each time is recorded as follows: Historical health monitoring data is used to build a GBDT (Gradient Boosting Decision Tree) model for subsequent sleep score prediction.
[0048] S2. Iterate through the preprocessed health monitoring data according to the GBDT algorithm to obtain the residual sequence of the data points corresponding to each preprocessed health monitoring data. Each time, all health monitoring data items collected constitute a data point.
[0049] It should be noted that during GBDT modeling using health monitoring data from smart wearable devices, there may be abnormal data in the collected health monitoring data, which may cause the residuals of the data points to be artificially high during the modeling process. The presence of abnormal data will lead to deviations in GBDT modeling. Therefore, this application reduces the residuals of each data point in the GBDT modeling process according to the probability of abnormality to obtain more accurate GBDT modeling.
[0050] In one embodiment, during the modeling of health monitoring data using the GBDT algorithm, abnormal data may fail to gradually decrease or decrease in a manner that does not conform to the big data patterns of health monitoring data. Therefore, this application uses the change sequence of the residuals of each data point during the GBDT modeling process to make the most intuitive judgment of anomalies in health monitoring data.
[0051] It should be noted that, during the GBDT algorithm modeling process, the residual changes for each data point are used to calculate the probability of anomalies. Specifically, the GBDT algorithm is used to perform initial modeling iterations on the sample health monitoring data, with a preset initial iteration count of [number missing]. , will the The data point at the th th The absolute value of the residual after the nth iteration is denoted as . No. The residual data of each data point are sorted according to the number of iterations to form the residual sequence of that data point. In this case, all health monitoring data items collected in each iteration constitute a data point.
[0052] S3. Based on the residual sequence, obtain the residual trend of each data point corresponding to the preprocessed health monitoring data.
[0053] In one embodiment, the residual trend of the data points is obtained based on the residual sequence of the data points:
[0054]
[0055] in, Indicates the first Residual trend of each data point and They represent the first The residual sequence of the nth data point The and the first One data point, Indicates the first The range of the residual sequence of data points.
[0056] Furthermore, in the formula This represents the ratio of the difference between two adjacent residual data points to the average difference. A larger cumulative product of these ratios indicates a more stable and normal variation in the residuals, and a better trend in the residuals. The formula contains... The partial value represents the difference between the residual data of two adjacent iterations. The larger the difference, the greater the degree of change in the residual and the smaller the trend of the residual.
[0057] At this point, the residual trend of all data points has been obtained.
[0058] S4. Based on the distribution of health monitoring data, obtain the probability of anomalies for each data point corresponding to each preprocessed health monitoring data.
[0059] It should be noted that for any health monitoring data point, if its probability of anomaly is higher compared to historically collected data, then that data point is considered likely to affect the GBDT modeling iteration. Similarly, if the residual variation of a data point is erratic, then that data point is also considered to have a high probability of anomaly. Secondly, for any data point, if its Euclidean distance to several recently collected data points is significantly greater than its Euclidean distance to other historically collected data points, it indicates that its data changes in recent periods may be abnormal. Based on this, the probability of anomaly for the data point can be calculated.
[0060] In one embodiment, a neighborhood data range is preset. For the first For each data point, the difference between its number of collections and the number of collections is less than or equal to [a certain value]. The other data points are denoted as its temporal neighbor data points, and the number of temporal neighbor data points is denoted as... , will the The first data point The Euclidean distance between a time-series neighborhood data point and its Euclidean distance is denoted as . , will be with the The data point with the nearest Euclidean distance The nth data point is denoted as its spatial neighborhood data point, and the nth data point is denoted as its spatial neighborhood data point. The data point and its first The Euclidean distance between 1 spatial neighborhood data points is denoted as . Based on this, the formula for calculating the probability of anomalies in data points is as follows:
[0061]
[0062] in, Indicates the first The probability of anomalies in each data point This represents the number of temporal and spatial neighboring data points of the i-th data point (the number of temporal data points and spatial data points are the same). Indicates the first The first data point The Euclidean distance between a time-series neighborhood data point and its Euclidean distance. Indicates the first The data point and its first Euclidean distance between spatial neighborhood data points.
[0063] Furthermore, in the formula Indicates the first The average Euclidean distance between the temporal neighbors of each data point and the data points in the neighborhood. Indicates the first The greater the difference between the average Euclidean distance of the spatial neighbors of the first data point and the second data point, the more indicative it is of the strength of the first data point. The time-series difference between the nearest other data points of the first data point is greater than that of the second data point. The presence of other data points in the neighborhood of a data point indicates that the difference between this data point and the other data points closest in time is large, suggesting that it may be an anomalous data point in time, meaning that the greater the probability of the data point being anomalous.
[0064] It should be noted that for any data point, if the residual change during the iteration process is not convergent, it indicates that there is a difference between this data point and all data points, meaning that this data point may be an outlier.
[0065] S5. Based on the probability of anomalies and the residual trend of each preprocessed health monitoring data point, obtain the corresponding residual reduction index.
[0066] It should be noted that for any given data point, the greater its probability of being an anomaly, the more likely it is to be an anomalous data point. In this case, its residual should be appropriately reduced to prevent it from affecting the algorithm. On the other hand, the smaller the trend of the residual of a data point, the more likely it is that the residual of that data point does not conform to the overall change pattern of the data. Similarly, its residual should also be appropriately reduced.
[0067] In one embodiment, the ratio of the residual trend to the probability of anomalies for each data point corresponding to the preprocessed health monitoring data is used as the residual reduction index for the data point. .
[0068] S6. The residuals of the preprocessed health monitoring data in the GBDT modeling process are weakened according to the residual reduction index.
[0069] In one embodiment, all data points whose residual reduction indices are less than a preset residual reduction index threshold are recorded as data points to be reduced; the residuals and residual reduction indices obtained by the data points to be reduced during the GBDT algorithm iteration process are compared. Multiply the results to obtain the reduced residuals; then, perform GBDT iteration based on the reduced residuals to obtain the corresponding prediction model.
[0070] As an example, a preset residual reduction exponential threshold is used. Obtain all residual reduction exponents that are less than the preset residual reduction exponent threshold. The data points are denoted as data points to be reduced. The residuals obtained in the subsequent GBDT algorithm iteration are multiplied by the residual reduction exponent to obtain the reduced residuals. GBDT iteration is then performed to obtain the corresponding prediction model, which can be used as the GBDT model after residual reduction.
[0071] S7. Monitor the sleep health status of smart wearable devices based on the GBDT model after residual reduction.
[0072] In one embodiment, the sleep score of a smart wearable device is predicted based on a GBDT model with weakened residuals; if the sleep score is lower than a preset sleep score threshold, it is determined that the human sleep is abnormal, and an alert is issued in a timely manner. The preset sleep score threshold is a pre-set threshold for determining whether a human sleep is abnormal, and this threshold can be the average of all historical sleep scores.
[0073] This application also discloses a health monitoring system for a smart wearable device, including:
[0074] The data acquisition module is used to collect health monitoring data from smart wearable devices and to preprocess the health monitoring data.
[0075] The residual sequence calculation module is used to iterate the preprocessed health monitoring data according to the GBDT algorithm to obtain the residual sequence of each data point corresponding to the preprocessed health monitoring data. All health monitoring data items collected in one session constitute a data point.
[0076] The residual trend calculation module is used to obtain the residual trend of each data point corresponding to the preprocessed health monitoring data based on the residual sequence.
[0077] The anomaly probability calculation module is used to obtain the anomaly probability of each data point corresponding to the preprocessed health monitoring data based on the distribution of the preprocessed health monitoring data.
[0078] The residual reduction index calculation module obtains the corresponding residual reduction index based on the probability of anomalies and the residual trend of each preprocessed health monitoring data point.
[0079] The residual reduction module is used to reduce the residuals of the preprocessed health monitoring data during GBDT modeling based on the residual reduction index.
[0080] The monitoring module is used to monitor the sleep health status of smart wearable devices based on the GBDT model after residual reduction.
[0081] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.
Claims
1. A health monitoring method for a smart wearable device, characterized in that, include: Collect health monitoring data from smart wearable devices and preprocess the health monitoring data; The preprocessed health monitoring data is iterated according to the GBDT algorithm to obtain the residual sequence of each data point corresponding to the preprocessed health monitoring data. Here, all health monitoring data items collected in one session constitute a data point. Based on the residual sequence, the residual trend of each data point corresponding to the preprocessed health monitoring data is obtained; Based on the distribution of the preprocessed health monitoring data, the probability of anomalies for each data point corresponding to the preprocessed health monitoring data is obtained. Based on the probability of anomalies and the residual trend of each data point corresponding to the preprocessed health monitoring data, the corresponding residual reduction index is obtained; The residuals of the preprocessed health monitoring data during GBDT modeling are weakened according to the residual reduction index. The sleep health status of smart wearable devices is monitored based on the GBDT model with residual weakening.
2. The health monitoring method for a smart wearable device according to claim 1, characterized in that, The process of collecting health monitoring data from smart wearable devices and preprocessing the health monitoring data includes: Collect sleep monitoring data from each session as health monitoring data; The first The first collection The data item is denoted as The total number of sleep data collections is denoted as n, and the number of data items collected in each sleep monitoring session is denoted as n. ; For each data point, a sleep score is calculated based on all health monitoring data. The score will be determined by the data point numbered... The sleep score corresponding to the data points collected each time is recorded as follows: GBDT modeling was performed using historical health monitoring data to predict subsequent sleep scores.
3. The health monitoring method for a smart wearable device according to claim 1, characterized in that, The step of iterating through the preprocessed health monitoring data according to the GBDT algorithm to obtain the residual sequence of each preprocessed health monitoring data includes: The GBDT algorithm is used to perform initial modeling iterations on the sample health monitoring data, where the preset initial iteration number is [number missing]. ; The first The data point at the th th The absolute value of the residual after the nth iteration is denoted as , among which, the The residual data of all data points are sorted according to the iteration number to form the... The residual sequence of data points.
4. The health monitoring method for a smart wearable device according to claim 1, characterized in that, The calculation formula corresponding to the probability of an anomaly includes: ， in, Indicates the first The probability of anomalies in each data point Indicates the first The number of temporal and spatial neighborhood data points for each data point Indicates the first The first data point The Euclidean distance between a time-series neighborhood data point and its Euclidean distance. Indicates the first The data point and its first Euclidean distance between spatial neighborhood data points Indicates the first The average Euclidean distance between time-series neighboring data points of each data point Indicates the first The average Euclidean distance between spatial neighbor data points of a given data point.
5. The health monitoring method for a smart wearable device according to claim 1, characterized in that, The calculation formula corresponding to the residual trend includes: ， in, Indicates the first Residual trend of each data point and They represent the first The residual sequence of the nth data point The and the first One data point, Indicates the first The range of the residual sequence of data points. This represents the ratio of the difference between two adjacent residual data points to the average difference. This represents the difference in residual data between two consecutive iterations.
6. The health monitoring method for a smart wearable device according to claim 1, characterized in that, The process of reducing the residuals of the preprocessed health monitoring data during GBDT modeling based on the residual reduction index includes: The ratio of the residual trend to the probability of anomalies of a data point is used as the residual reduction index for that data point. All data points whose residual reduction index is less than the preset residual reduction index threshold are recorded as data points to be reduced; The residual obtained during the GBDT algorithm iteration process of the data points to be reduced is multiplied by the residual reduction exponent to obtain the reduced residual; Based on the reduced residuals, GBDT iteration is performed to obtain the corresponding prediction model.
7. The health monitoring method for a smart wearable device according to claim 1, characterized in that, The monitoring of sleep health status of smart wearable devices based on the residual-weakened GBDT model includes: Predict sleep scores for smart wearable devices based on the residual-weakened GBDT model; If the sleep score is lower than the preset sleep score threshold, it is determined that the human sleep is abnormal and a timely reminder is issued.
8. A health monitoring system for a smart wearable device, characterized in that, include: The data acquisition module is used to collect health monitoring data from smart wearable devices and to preprocess the health monitoring data. The residual sequence calculation module is used to iterate the preprocessed health monitoring data according to the GBDT algorithm to obtain the residual sequence of each data point corresponding to the preprocessed health monitoring data, wherein all health monitoring data items collected in one time constitute a data point; The residual trend calculation module is used to obtain the residual trend of each data point corresponding to the preprocessed health monitoring data based on the residual sequence. An anomaly probability calculation module is used to obtain the anomaly probability of each data point corresponding to the preprocessed health monitoring data based on the distribution of the preprocessed health monitoring data. The residual reduction index calculation module obtains the corresponding residual reduction index based on the probability of anomalies and the residual trend of each data point corresponding to the preprocessed health monitoring data. The residual reduction module is used to reduce the residuals of the preprocessed health monitoring data during the GBDT modeling process according to the residual reduction index. The monitoring module is used to monitor the sleep health status of smart wearable devices based on the GBDT model after residual reduction.