Health monitoring-based alarm method, device, equipment and storage medium

Abstract

The application discloses a health monitoring-based alarm method and device, equipment and a storage medium. The method comprises: acquiring at least one monitoring data representing a physiological index of a user; constructing current sample data based on the at least one monitoring data; determining an abnormal score value of the current sample data based on the current sample data and a historical sample data set by using a robust random forest algorithm; and generating alarm information if the abnormal score value is outside a numerical interval of an abnormal score threshold value. In this way, intelligent alarm can be performed based on a comparison result of the abnormal score value of the current sample data and the abnormal score threshold value of the historical sample data set. The abnormal score threshold value can be updated based on the update of the historical sample data set of the user, thereby providing individualized and accurate intelligent alarm for the user, effectively improving the accuracy of health monitoring alarm, reducing the occurrence of false alarms, and further improving the efficiency of medical resource allocation.

Description

Technical Field

[0001] This application relates to the field of health monitoring, and more particularly to an alarm method, device, equipment, and storage medium based on health monitoring. Background Technology

[0002] With the aging population becoming increasingly serious, the need for health monitoring among the elderly is becoming more and more urgent. Because the elderly often have multiple chronic diseases, with long disease courses and complex causes, leading to increased disability or mortality rates, dynamic monitoring and management of chronic diseases are crucial.

[0003] In related technologies, various monitoring devices exist, such as smart bracelets and portable monitors, which can monitor relevant physiological indicators and trigger alarms based on the monitored data. However, related emergency alarm systems mainly collect monitoring data from the elderly and automatically trigger emergency calls based on whether the monitored data is abnormal; or they may trigger emergency calls based on the elderly's subjective feeling of discomfort. For example, if the elderly person's data reaches the peak value set in the system's daily data collection, or if the elderly person feels their safety is threatened and triggers an alarm voluntarily. However, the elderly person's condition may change, and relying on previously set peak values ​​to determine whether an emergency call is needed may lead to untimely emergency calls or frequent false alarms. On the other hand, the elderly person's lack of understanding of their own condition or loneliness may also lead to false alarms. Such false alarms pose a significant challenge to the healthcare system and affect the rational allocation of medical resources. Summary of the Invention

[0004] In view of this, embodiments of this application provide an alarm method, apparatus, device, and storage medium based on health monitoring, aiming to improve the alarm accuracy of health monitoring and reduce the false alarm rate.

[0005] The technical solution of this application embodiment is implemented as follows:

[0006] In a first aspect, embodiments of this application provide an alarm method based on health monitoring, comprising:

[0007] Acquire at least one monitoring data point representing a user's physiological indicators;

[0008] The current sample data is constructed based on at least one of the monitoring data;

[0009] Based on the current sample data and historical sample dataset, a robust random segmentation forest algorithm is used to determine the anomaly score value of the current sample data;

[0010] If the abnormal score value is determined to be outside the range of the abnormal score threshold, an alarm message is generated.

[0011] Wherein, the number of data dimensions of each sample data in the current sample data and the historical sample dataset is d, where d is a natural number greater than or equal to 1, and the anomaly scoring threshold is determined based on the anomaly scoring value of each sample data in the historical sample dataset.

[0012] In the above scheme, the step of determining the anomaly score value of the current sample data using a robust random segmentation forest algorithm based on the current sample data and historical sample dataset includes:

[0013] A first binary tree is constructed based on the first algorithm for the historical sample dataset;

[0014] A second binary tree is constructed based on the second algorithm for the merged sample dataset of the historical sample dataset and the current sample data;

[0015] For each sample data in the historical sample dataset, the sum of the depth differences between the second binary tree and the first binary tree for each sample data is calculated to obtain the anomaly score value of the current sample data.

[0016] In the above scheme, constructing a first binary tree based on the historical sample dataset using the first algorithm includes:

[0017] Calculate the maximum and minimum values ​​for each data dimension of the historical sample dataset;

[0018] A first segmentation threshold is determined for a first target dimension used for binary tree partitioning, wherein the first target dimension is any one of d data dimensions, and the first segmentation threshold is less than the maximum value of the first target dimension and greater than the minimum value of the first target dimension.

[0019] The historical sample dataset, which serves as the root node, is divided into a binary tree based on the first segmentation threshold. The first segmentation threshold of each node after the division is redefined. The binary tree is then divided into two parts based on the corresponding first segmentation threshold until the division is complete, thus obtaining the first binary tree.

[0020] In the above scheme, the step of constructing a second binary tree based on the merged sample dataset of the historical sample dataset and the current sample data using the second algorithm includes:

[0021] The maximum, minimum, and range of the merged sample dataset in each data dimension are calculated.

[0022] Determine a baseline value, wherein the value of the baseline value is greater than zero and less than the sum of the ranges of each data dimension;

[0023] A second segmentation threshold is determined for the second target dimension used for binary tree partitioning. The second target dimension is the data dimension with the smallest index among d data dimensions whose cumulative range sum is greater than or equal to the benchmark value. The cumulative range sum is the sum of the ranges of the second target dimension and all data dimensions before the second target dimension, counted in the order of the data dimensions. The second segmentation threshold is the minimum value of the second target dimension plus the cumulative range sum minus the benchmark value.

[0024] Compare the second segmentation threshold with the maximum and minimum values ​​of the second target dimension statistically obtained from the historical sample dataset in the first algorithm;

[0025] If the second segmentation threshold is greater than or equal to the maximum value of the second target dimension of the historical sample dataset in the first algorithm, or less than or equal to the minimum value of the second target dimension of the historical sample dataset in the first algorithm, then a second binary tree is generated based on the comparison result of the value of the second target dimension of the current sample data with the second segmentation threshold and the first binary tree. The second binary tree includes the current sample data and the first binary tree located at the next level below the root node. If the value of the second target dimension of the current sample data is greater than the second segmentation threshold, then the current sample data is located on the right side of the first binary tree. If the value of the second target dimension of the current sample data is less than or equal to the second segmentation threshold, then the current sample data is located on the left side of the first binary tree.

[0026] If the second segmentation threshold is less than the maximum value of the second target dimension of the historical sample dataset as counted in the first algorithm and greater than the minimum value of the second target dimension of the historical sample dataset as counted in the first algorithm, then the merged sample dataset, which serves as the root node, is partitioned into a binary tree based on the first algorithm to obtain the second binary tree.

[0027] The method in the above scheme further includes:

[0028] Construct the initial sample dataset for the user;

[0029] The anomaly score value of each sample data in the initial sample dataset is obtained based on the first algorithm and the second algorithm;

[0030] The set range of the score distribution of the abnormal score values ​​of each sample data in the initial sample set is used as the initial abnormal score threshold for the user.

[0031] The initial sample set data serves as the historical sample dataset for the user's first health monitoring, and the initial anomaly score threshold serves as the anomaly score threshold for the user's first health monitoring.

[0032] In the above scheme, after determining the anomaly score value of the current sample data using a robust random segmentation forest algorithm based on the current sample data and historical sample dataset, the method further includes:

[0033] The current sample dataset is added to the historical sample dataset to serve as the historical sample dataset for the user's next health monitoring.

[0034] Based on the first algorithm and the second algorithm, the abnormal score value of each sample data in the historical sample dataset of the user's next health monitoring is obtained, and the abnormal score threshold of the user's next health monitoring is updated.

[0035] The method in the above scheme further includes:

[0036] The user's characteristic data is obtained, which includes at least one of the following: first attribute data on whether the user is taking medication, second attribute data on whether the user suffers from the target disease, and third attribute data on the measurement method of the target physiological indicator.

[0037] The data dimensions of the user's sample data are determined based on the feature data.

[0038] Secondly, embodiments of this application provide an alarm device based on health monitoring, comprising:

[0039] The acquisition module is used to acquire at least one monitoring data point representing the user's physiological indicators;

[0040] The sample construction module is used to construct the current sample data based on the at least one monitoring data.

[0041] The scoring module is used to determine the outlier scores of the current sample data based on the current sample data and the historical sample dataset using a robust random segmentation forest algorithm.

[0042] An alarm module is used to generate an alarm message if the abnormal score value is outside the numerical range of the abnormal score threshold.

[0043] Wherein, the number of data dimensions of each sample data in the current sample data and the historical sample dataset is d, where d is a natural number greater than or equal to 1, and the anomaly scoring threshold is determined based on the anomaly scoring value of each sample data in the historical sample dataset.

[0044] Thirdly, embodiments of this application provide an electronic device, including: a processor and a memory for storing a computer program capable of running on the processor, wherein, when the processor is used to run the computer program, it executes the steps of the method described in the first aspect of embodiments of this application.

[0045] Fourthly, embodiments of this application provide a computer storage medium storing a computer program, which, when executed by a processor, implements the steps of the method described in the first aspect of embodiments of this application.

[0046] The technical solution provided in this application involves acquiring at least one monitoring data point representing a user's physiological indicators; constructing current sample data based on the at least one monitoring data point; determining the abnormal score value of the current sample data using a robust random segmentation forest algorithm based on the current sample data and historical sample datasets; and generating an alarm message if the abnormal score value is determined to be a value outside the numerical range of the abnormal score threshold. Here, the number of data dimensions for each sample data point in both the current sample data and the historical sample dataset is 'd', where 'd' is a natural number greater than or equal to 1, and the abnormal score threshold is determined based on the abnormal score values ​​of each sample data point in the historical sample dataset. Thus, intelligent alarms can be generated based on the comparison between the abnormal score value of the current sample data and the abnormal score threshold of the historical sample dataset. This abnormal score threshold can be updated based on updates to the user's historical sample dataset, providing users with personalized and precise intelligent alarms, effectively improving the accuracy of health monitoring alarms, reducing false alarms, and thereby improving the efficiency of medical resource allocation. Attached Figure Description

[0047] Figure 1 This is a flowchart illustrating the alarm method based on health monitoring, as described in an embodiment of this application.

[0048] Figure 2 This is a schematic diagram illustrating the process of building, applying, and updating the monitoring model in an application example of this application;

[0049] Figure 3 This is a schematic diagram of the structure of an alarm device based on health monitoring according to an embodiment of this application;

[0050] Figure 4 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0051] The present application will now be described in further detail with reference to the accompanying drawings and embodiments.

[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0053] This application provides an alarm method based on health monitoring. This alarm method can be applied to electronic devices with data processing capabilities, such as terminal devices and servers, or it can be implemented through the cooperation of a terminal device and a server. Specifically, the terminal device can be a computer, smartphone, personal digital assistant (PDA), etc.; the server can be an application server or a web server. In actual deployment, the server can be a standalone server or a cluster server. Figure 1 As shown, the alarm method includes:

[0054] Step 101: Obtain at least one monitoring data point representing the user's physiological indicators.

[0055] For example, the electronic device can acquire at least one monitoring data of the user, such as blood pressure data, blood glucose data, and heart rate data.

[0056] Here, the electronic device can communicate with various monitoring devices, or the electronic device can have built-in monitoring functions for the aforementioned monitoring data. This application embodiment does not limit this.

[0057] Step 102: Construct the current sample data based on the at least one monitoring data.

[0058] It should be noted that in this embodiment of the application, the data dimension of the user's sample data is uniform. For example, the number of data dimensions of the current sample data and each historical sample data is d, where d is a natural number greater than or equal to 1.

[0059] For example, in embodiments of this application, before constructing a monitoring model for a user, the user's feature data can be obtained. The feature data includes, but is not limited to, at least one of the following: first attribute data on whether the user is taking medication, second attribute data on whether the user suffers from a target disease, and third attribute data on the measurement method of the target physiological indicator; the data dimension of the user's sample data can be determined based on the feature data.

[0060] It should be noted that in related technologies, alarms based on health monitoring do not take into account the impact of changes in the elderly's condition, medication, and measurement methods, which can lead to false alarms. Therefore, in this embodiment, by acquiring the user's characteristic data and determining the required data dimensions based on that data, health monitoring alarms can be implemented more scientifically and reasonably.

[0061] It's important to note that user characteristic data doesn't function as data dimensions, but rather determines the corresponding data dimensions for each user. For example, if the first attribute data is "yes," then blood pressure data can be divided into multiple data dimensions, such as blood pressure data before medication and blood pressure data after medication for a set period of time. If the second attribute data indicates that the user has diabetes, then blood glucose data can be divided into data dimensions such as fasting blood glucose value and blood glucose data after a set period of time. If the third attribute data indicates that the user measures blood pressure on their arm or wrist, then blood pressure data can be divided into data dimensions such as arm measurement value and wrist measurement value.

[0062] In one application example, assuming the user is an elderly person with hypertension and diabetes who takes medication long-term, the data dimensions of this user's sample data are shown in Table 1 below:

[0063] Table 1

[0064]

[0065] It should be noted that if a user's characteristic data changes, the data dimensions of the user's sample data will change, requiring the monitoring model for that user to be rebuilt.

[0066] For example, the current sample data is constructed based on the at least one monitoring data, including:

[0067] If at least one of the acquired monitoring data has a missing data dimension, then the missing data dimension is filled based on the corresponding value in the previous sample data.

[0068] Here, each sample data can be a sample sequence based on monitoring time. Electronic devices can construct a sample data based on monitoring data of all data dimensions, or they can construct a sample data based on partial monitoring data. This application embodiment does not limit this.

[0069] Preferably, the electronic device can generate new sample data when at least one monitoring data is updated, thereby effectively reducing alarm delay and improving alarm timeliness while ensuring alarm accuracy.

[0070] Taking the sample data shown in Table 1 as an example, if the current data obtained is "the user's blood pressure one hour after taking the medication", then the current sample data can be constructed based on the obtained "blood pressure one hour after taking the medication" and combined with the previous sample data "blood pressure before taking the medication", "fasting blood glucose", "blood glucose 2 hours after a meal" and "resting heart rate".

[0071] Step 103: Based on the current sample data and the historical sample dataset, the robust random split forest algorithm is used to determine the anomaly score value of the current sample data.

[0072] Here, after constructing the current sample data, the electronic device can use the Robust Random Cut Forest (RRCF) algorithm to determine the anomaly score of the current sample data based on the current sample data and the historical sample dataset, and issue an alarm based on the comparison result of the anomaly score and the anomaly score threshold, which is determined based on the anomaly score of each sample data in the historical sample dataset.

[0073] Step 104: If the abnormal score value is determined to be outside the numerical range of the abnormal score threshold, then an alarm message is generated.

[0074] Here, based on the comparison between the abnormal score value and the abnormal score threshold of the current sample data, if the abnormal score value is less than or equal to the lower limit of the abnormal score threshold or greater than or equal to the upper limit of the abnormal score threshold, the electronic device determines that the abnormal score value is outside the numerical range of the abnormal score threshold and can generate alarm information to realize intelligent alarm for user health monitoring.

[0075] For example, the generated alarm information may be to send a call request to the call center of the medical system to establish a voice call, or to send a warning message carrying information such as the user's home address to the call platform of the medical system. This application embodiment does not limit this.

[0076] It is understood that the alarm method in this application embodiment can make intelligent alarms based on the comparison results of the abnormal score value of the current sample data and the abnormal score threshold of the historical sample dataset. The abnormal score threshold can be updated based on the update of the user's historical sample dataset, which can provide users with personalized and accurate intelligent alarms, effectively improve the accuracy of health monitoring alarms, reduce the occurrence of false alarms, and thus improve the efficiency of medical resource allocation.

[0077] For example, determining the anomaly score of the current sample data using a robust random segmentation forest algorithm based on the current sample data and historical sample dataset includes:

[0078] A first binary tree is constructed based on the first algorithm for the historical sample dataset;

[0079] A second binary tree is constructed based on the second algorithm for the merged sample dataset of the historical sample dataset and the current sample data;

[0080] For each sample data in the historical sample dataset, the sum of the depth differences between the second binary tree and the first binary tree for each sample data is calculated to obtain the anomaly score value of the current sample data.

[0081] It is understandable that each sample data in the historical sample dataset serves as a leaf node in the aforementioned first and second binary trees. For each sample data, the depth difference between the second and first binary trees can be calculated. For example, if the depth of sample data Xi in the second binary tree (i.e., the level number in the binary tree, which increases sequentially from the root node) is D2i, and the depth of sample data Xi in the first binary tree is D1i, then the depth difference of sample data Xi is D2i-D1i. By summing the depth differences of each sample data in the historical sample dataset, the abnormal score value of the current sample data can be obtained.

[0082] For example, constructing a first binary tree on the historical sample dataset based on the first algorithm includes:

[0083] Calculate the maximum and minimum values ​​for each data dimension of the historical sample dataset;

[0084] A first segmentation threshold is determined for a first target dimension used for binary tree partitioning, wherein the first target dimension is any one of d data dimensions, and the first segmentation threshold is less than the maximum value of the first target dimension and greater than the minimum value of the first target dimension.

[0085] The historical sample dataset, which serves as the root node, is divided into a binary tree based on the first segmentation threshold. The first segmentation threshold of each node after the division is redefined. The binary tree is then divided into two parts based on the corresponding first segmentation threshold until the division is complete, thus obtaining the first binary tree.

[0086] Understandably, for all sample data in the historical sample dataset, the maximum and minimum values ​​of each data dimension in d data dimensions are calculated. For the root node (i.e., the historical sample dataset), data dimension i is randomly selected from the d data dimensions as the first target dimension, and the corresponding first splitting threshold C is between the maximum and minimum values ​​of data dimension i. For any sample data, if the value of the sample data in data dimension i is less than or equal to the first splitting threshold C, it enters the left node; otherwise, it enters the right node. Then, for the left and right nodes under the root node, the first splitting threshold of the first target dimension for binary tree partitioning is determined (the process is similar to that of the root node), and the data samples of the corresponding nodes are partitioned into a binary tree. This process is repeated until the partitioning is complete, resulting in the first binary tree.

[0087] For example, constructing a second binary tree based on the merged sample dataset of the historical sample dataset and the current sample data using the second algorithm includes:

[0088] The maximum, minimum, and range of the merged sample dataset in each data dimension are calculated.

[0089] Determine a baseline value, wherein the value of the baseline value is greater than zero and less than the sum of the ranges of each data dimension;

[0090] A second segmentation threshold is determined for the second target dimension used for binary tree partitioning. The second target dimension is the data dimension with the smallest index among d data dimensions whose cumulative range sum is greater than or equal to the benchmark value. The cumulative range sum is the sum of the ranges of the second target dimension and all data dimensions before the second target dimension, counted in the order of the data dimensions. The second segmentation threshold is the minimum value of the second target dimension plus the cumulative range sum minus the benchmark value.

[0091] Compare the second segmentation threshold with the maximum and minimum values ​​of the second target dimension statistically obtained from the historical sample dataset in the first algorithm;

[0092] If the second segmentation threshold is greater than or equal to the maximum value of the second target dimension of the historical sample dataset in the first algorithm, or less than or equal to the minimum value of the second target dimension of the historical sample dataset in the first algorithm, then a second binary tree is generated based on the comparison result of the value of the second target dimension of the current sample data with the second segmentation threshold and the first binary tree. The second binary tree includes the current sample data and the first binary tree located at the next level below the root node. If the value of the second target dimension of the current sample data is greater than the second segmentation threshold, then the current sample data is located on the right side of the first binary tree. If the value of the second target dimension of the current sample data is less than or equal to the second segmentation threshold, then the current sample data is located on the left side of the first binary tree.

[0093] If the second segmentation threshold is less than the maximum value of the second target dimension of the historical sample dataset as counted in the first algorithm and greater than the minimum value of the second target dimension of the historical sample dataset as counted in the first algorithm, then the merged sample dataset, which serves as the root node, is partitioned into a binary tree based on the first algorithm to obtain the second binary tree.

[0094] Understandably, through the processing of the second algorithm described above, a second binary tree can be constructed from the merged historical sample dataset and the current sample dataset. Thus, based on the first and second binary trees, the anomaly score of the current sample data can be obtained.

[0095] It should be noted that the alarm method in this application embodiment also needs to include a training phase for the initial monitoring model, which may specifically include:

[0096] Construct the initial sample dataset for the user;

[0097] The anomaly score value of each sample data in the initial sample dataset is obtained based on the first algorithm and the second algorithm;

[0098] The set range of the score distribution of the abnormal score values ​​of each sample data in the initial sample set is used as the initial abnormal score threshold for the user.

[0099] The initial sample set data serves as the historical sample dataset for the user's first health monitoring, and the initial anomaly score threshold serves as the anomaly score threshold for the user's first health monitoring.

[0100] For example, before building the initial monitoring model, user feature data can be obtained to determine the data dimensions of the user's sample data. Then, information such as big data from the network and / or the user's historical monitoring data can be used to build the user's initial sample dataset. The number of samples in the initial sample dataset can be reasonably set based on the requirements.

[0101] For example, assuming the initial sample dataset has n samples, one sample can be selected, and the remaining n-1 samples form a dataset. A first binary tree is constructed on this dataset using the first algorithm mentioned above. A second binary tree is constructed on the dataset after the selected sample is added, using the second algorithm mentioned above. The anomaly score of the sample is determined based on the difference between the first and second binary trees. This process is repeated to obtain the anomaly scores of each sample in the initial sample dataset. The distribution of the anomaly scores of the n samples is then calculated, and the 95% range in descending order (i.e., the range of 2.5% to 97.5% of the score distribution) is taken as the initial anomaly score threshold range.

[0102] For example, after determining the anomaly score value of the current sample data using a robust random segmentation forest algorithm based on the current sample data and historical sample dataset, the method further includes:

[0103] The current sample dataset is added to the historical sample dataset to serve as the historical sample dataset for the user's next health monitoring.

[0104] Based on the first algorithm and the second algorithm, the abnormal score value of each sample data in the historical sample dataset of the user's next health monitoring is obtained, and the abnormal score threshold of the user's next health monitoring is updated.

[0105] Understandably, the user's monitoring model can be updated. For example, newly acquired current sample data can be added to the historical sample dataset to obtain an updated historical sample dataset. Based on the aforementioned first and second algorithms, the anomaly score value for each sample in the updated historical sample dataset can be calculated. The distribution of anomaly scores across all sample data can be statistically analyzed, and the 95th percentile range (i.e., the range from 2.5% to 97.5% of the score distribution) after descending order is taken as the updated anomaly score threshold range. In this way, the anomaly score threshold can be dynamically adjusted in real time based on health monitoring data, achieving adaptive adjustment of the anomaly score threshold, thereby improving the accuracy of alarms and reducing false alarms.

[0106] The following application example will be used to further describe this application in detail.

[0107] Reference Figure 2 In this application example, the alarm method based on health monitoring includes a model training phase and a model evaluation phase. The model training phase trains the RRCF model based on the sample dataset to obtain an anomaly score threshold that serves as the alarm evaluation benchmark. The model evaluation phase involves preprocessing the user's monitoring data (i.e., constructing the current sample data), obtaining the current anomaly score value based on RRCF anomaly detection, and then generating an intelligent alarm based on a comparison between the current anomaly score value and the anomaly score threshold.

[0108] The training and evaluation phases of the model are explained below:

[0109] I. RRCF Model Training Phase

[0110] This application example illustrates the training of the initial RRCF model, which may include the following steps:

[0111] Step 1: Collect users' health monitoring data and build an initial sample dataset.

[0112] Here, health monitoring data for elderly individuals during non-acute illness periods can be collected (to avoid the influence of acute illness on health monitoring indicators). This data is divided into two parts: monitoring data (including but not limited to blood pressure, blood sugar, and heart rate) and characteristic data (such as measurement time, measurement site, medication status, etc.). The measurement site affects blood pressure values; typically, blood pressure is measured on the arm or wrist, and is represented using binary values, such as 0 for arm blood pressure and 1 for wrist blood pressure. Medication status refers to whether the elderly individual needs to take medication, represented using binary data of 0 and 1, where 1 indicates medication is needed and 0 indicates no medication is needed.

[0113] Based on the collection time, the data is organized into time series data, resulting in an initial sample dataset S. Each time point is considered a sample point, and there are n sample points. Each sample point (i.e., sample data) is represented by X, and each sample point X includes d dimensions. The number of dimensions is determined by the feature data, but the feature data is not counted as dimensions. Each physiological indicator may correspond to multiple dimensions (e.g., blood pressure has pre-medication blood pressure and blood pressure 1 hour after medication). Monitoring data can be represented by x.

[0114] Through data preprocessing, the sample data of all sample points can be transformed into an n-row, d-column matrix, as shown below:

[0115]

[0116] Where, x 11 For the monitoring data of the first data dimension of the first sample point, x 12 For the monitoring data of the second data dimension of the first sample point, x 21The monitoring data for the first data dimension of the second sample point, and so on.

[0117] Step 2: Determine the anomaly score for each sample in the initial sample dataset based on the RRCF algorithm.

[0118] The basic principle of the RRCF algorithm is to maintain a binary tree structure generated by a sliding window of time series data, where each leaf node is a sample point and the root node is the original data set.

[0119] For an initial sample dataset S containing n samples, the first n-1 sample points are selected as the monitoring sample dataset S' to initialize the RRCF model. At this time, the RRCF model is a binary tree (i.e., the first binary tree) containing n-1 leaf nodes.

[0120] The first binary tree can be constructed using the first algorithm, as follows:

[0121] Step 1), calculate the maximum value x of each data dimension of all sample data in the monitoring sample dataset S'. max Minimum value x min ;

[0122] Step 2), randomly select the segmentation dimension i (i.e., the first target dimension);

[0123] Step 3) Determine the segmentation value C (i.e., the first segmentation threshold) for segmentation dimension i. The segmentation value C is the maximum value x in segmentation dimension i. max and minimum value x min between;

[0124] Step 4): For any sample data, if its value in the splitting dimension i is less than or equal to the splitting value C, then proceed to the left node; otherwise, proceed to the right node.

[0125] Step 5) Recursively generate the left and right subtrees until the partitioning is complete, and obtain the first binary tree.

[0126] After constructing the first binary tree, a sample data X' (e.g., the nth sample point) from the initial sample dataset S, other than the previously constructed monitoring sample dataset S', can be input, and a second binary tree can be constructed based on the second algorithm, as follows:

[0127] Step 1), calculate the maximum value x' of each data dimension of all sample data in the initial sample dataset S. max Minimum value x' min and range l';

[0128] Step 2), randomly select parameter r (i.e., the aforementioned baseline value), where r ranges from 0 to... The values ​​between;

[0129] Step 3), select the segmentation dimension i (i.e., the second target dimension) such that (The sum of the ranges from the first dimension to the i-th dimension) ≥ r and Take the minimum value;

[0130] Step 4), calculate the segmentation value C' (i.e., the second segmentation threshold), C' = x' min,i (Minimum value of the i-th dimension)

[0131] Step 5): If the segmentation value C' is not applicable to the segmentation of the RRCF tree constructed by the original first algorithm (i.e., the segmentation value C' exceeds the maximum and minimum values ​​of dimension i in the original monitoring sample dataset S'), a new root node u (containing all sample points of dataset S) is generated. If the value of dimension i of sample data X' is greater than the segmentation value C', then sample data X' becomes the right subtree of the new root node u, and the original RRCF tree (i.e., the first binary tree) becomes the left subtree of the new root node u; if the value of dimension i of sample data X' is less than or equal to the segmentation value C', then sample data X' becomes the left subtree of the new root node u, and the original RRCF tree becomes the right subtree of the new root node u, thus obtaining the second binary tree.

[0132] Step 6): If the splitting value C' is suitable for the splitting of the RRCF tree constructed by the original first algorithm, then take the root node of the RRCF tree and use the original splitting value C of that node to perform binary tree partitioning. If the value of dimension i of the sample data X' is less than or equal to the value of C, then enter the left node; otherwise, enter the right node. That is, use the first algorithm to perform binary tree partitioning to obtain the second binary tree.

[0133] It should be noted that the outlier score of each sample point can be understood as the degree to which the inclusion or exclusion of that sample point changes the overall data structure, i.e., the difference between the first binary tree and the second binary tree.

[0134] For example, the anomaly score S DC The calculation formula is as follows:

[0135] S DC =∑ y∈S′ [f(y,ST)-f(y,S′T)]

[0136] Where y refers to a sample point belonging to the monitoring sample dataset S'. f refers to the depth of the leaf node corresponding to a sample point y in the RRCF tree. ST is the RRCF tree (i.e., the second binary tree) constructed by the second algorithm, and S′T is the RRCF tree (i.e., the first binary tree) constructed by the first algorithm. The calculation method of this formula is as follows: for the leaf nodes corresponding to all sample points in the monitoring sample dataset S', calculate the sum of the differences between the depths of all leaf nodes in the RRCF trees constructed by the second algorithm and the first algorithm to obtain the anomaly score value. It should be noted that the closer the anomaly score value is to 0, the lower the probability that the sample point is an anomalous sample point.

[0137] It is understandable that by traversing each sample point in the initial sample dataset S and repeating the processing of the first and second algorithms mentioned above, the anomaly score value of each sample point can be obtained.

[0138] Step 3: Determine the anomaly scoring threshold.

[0139] For example, the abnormal scores assigned to all sample points can be sorted in descending order, and the range of 2.5% to 97.5% of the score distribution can be taken as the normal value (medical reference range), and the rest can be abnormal values.

[0140] II. Model Evaluation Phase

[0141] It is understandable that, during the model evaluation phase, intelligent alarms can be triggered based on health monitoring data and anomaly scoring thresholds obtained during the model training phase. This can include the following steps:

[0142] Step 1: Obtain user monitoring data and perform data preprocessing.

[0143] Here, the latest monitoring data of the user can be obtained, and the current sample data can be constructed based on the obtained monitoring data. If there are missing data dimensions in the obtained monitoring data, the missing data dimensions are filled in based on the corresponding values ​​in the previous sample data.

[0144] Step 2: RRCF outlier check to obtain anomaly score values.

[0145] Here, the first algorithm can be repeated based on the historical sample dataset to obtain the RRCF tree, and then the second algorithm can be repeated on the newly generated sample points to obtain the RRCF tree, and the anomaly score value of the newly generated sample points can be calculated.

[0146] Step 3: Compare the abnormal score with the abnormal score threshold.

[0147] Here, if the abnormal score value is less than or equal to the lower limit of the abnormal score threshold or greater than or equal to the upper limit of the abnormal score threshold, the abnormal score value is determined to be an abnormal value, and an emergency call alarm is automatically triggered; if the abnormal score value falls within the range of the abnormal score threshold, the abnormal score value is determined to be a normal value, and health monitoring continues.

[0148] It should be noted that the newly generated sample points will be added to the monitoring dataset to form the updated historical sample dataset, which will then be used as the dataset for the next model training, thereby updating the anomaly scoring threshold.

[0149] Understandably, as the number of sample points in the monitoring dataset increases, the updated anomaly scoring threshold becomes increasingly aligned with the user's individual characteristics, providing personalized and precise intelligent alarms, effectively improving the accuracy of health monitoring alarms, reducing false alarms, and thus improving the efficiency of medical resource allocation.

[0150] In order to implement the method of the embodiments of this application, the embodiments of this application also provide an alarm device based on health monitoring. The alarm device based on health monitoring corresponds to the alarm method based on health monitoring described above, and the steps in the embodiments of the alarm method based on health monitoring are also fully applicable to the embodiments of the alarm device based on health monitoring.

[0151] like Figure 3 As shown, the alarm device based on health monitoring includes: an acquisition module 301, a sample construction module 302, a scoring module 303, and an alarm module 304. The acquisition module 301 is used to acquire at least one monitoring data representing the user's physiological indicators; the sample construction module 302 is used to construct current sample data based on the at least one monitoring data; the scoring module 303 is used to determine the abnormal score value of the current sample data using a robust random segmentation forest algorithm based on the current sample data and historical sample datasets; the alarm module 304 is used to generate an alarm message if the abnormal score value is outside the numerical range of the abnormal score threshold. Wherein, the number of data dimensions for each sample data in the current sample data and the historical sample dataset is d, where d is a natural number greater than or equal to 1, and the abnormal score threshold is determined based on the abnormal score values ​​of each sample data in the historical sample dataset.

[0152] In some embodiments, the scoring module 303 is specifically used for:

[0153] A first binary tree is constructed based on the first algorithm for the historical sample dataset;

[0154] A second binary tree is constructed based on the second algorithm for the merged sample dataset of the historical sample dataset and the current sample data;

[0155] For each sample data in the historical sample dataset, the sum of the depth differences between the second binary tree and the first binary tree for each sample data is calculated to obtain the anomaly score value of the current sample data.

[0156] In some embodiments, the scoring module 303 constructs a first binary tree on the historical sample dataset based on a first algorithm, including:

[0157] Calculate the maximum and minimum values ​​for each data dimension of the historical sample dataset;

[0158] A first segmentation threshold is determined for a first target dimension used for binary tree partitioning, wherein the first target dimension is any one of d data dimensions, and the first segmentation threshold is less than the maximum value of the first target dimension and greater than the minimum value of the first target dimension.

[0159] The historical sample dataset, which serves as the root node, is divided into a binary tree based on the first segmentation threshold. The first segmentation threshold of each node after the division is redefined. The binary tree is then divided into two parts based on the corresponding first segmentation threshold until the division is complete, thus obtaining the first binary tree.

[0160] In some embodiments, the scoring module 303 constructs a second binary tree based on the merged sample dataset of the historical sample dataset and the current sample data using a second algorithm, including:

[0161] The maximum, minimum, and range of the merged sample dataset in each data dimension are calculated.

[0162] Determine a baseline value, wherein the value of the baseline value is greater than zero and less than the sum of the ranges of each data dimension;

[0163] A second segmentation threshold is determined for the second target dimension used for binary tree partitioning. The second target dimension is the data dimension with the smallest index among d data dimensions whose cumulative range sum is greater than or equal to the benchmark value. The cumulative range sum is the sum of the ranges of the second target dimension and all data dimensions before the second target dimension, counted in the order of the data dimensions. The second segmentation threshold is the minimum value of the second target dimension plus the cumulative range sum minus the benchmark value.

[0164] Compare the second segmentation threshold with the maximum and minimum values ​​of the second target dimension statistically obtained from the historical sample dataset in the first algorithm;

[0165] If the second segmentation threshold is greater than or equal to the maximum value of the second target dimension of the historical sample dataset in the first algorithm, or less than or equal to the minimum value of the second target dimension of the historical sample dataset in the first algorithm, then a second binary tree is generated based on the comparison result of the value of the second target dimension of the current sample data with the second segmentation threshold and the first binary tree. The second binary tree includes the current sample data and the first binary tree located at the next level below the root node. If the value of the second target dimension of the current sample data is greater than the second segmentation threshold, then the current sample data is located on the right side of the first binary tree. If the value of the second target dimension of the current sample data is less than or equal to the second segmentation threshold, then the current sample data is located on the left side of the first binary tree.

[0166] If the second segmentation threshold is less than the maximum value of the second target dimension of the historical sample dataset as counted in the first algorithm and greater than the minimum value of the second target dimension of the historical sample dataset as counted in the first algorithm, then the merged sample dataset, which serves as the root node, is partitioned into a binary tree based on the first algorithm to obtain the second binary tree.

[0167] In some embodiments, the alarm device based on health monitoring further includes: a model training module 305, used to construct an initial sample dataset for the user; to obtain the abnormal score value of each sample data in the initial sample dataset based on the first algorithm and the second algorithm; and to statistically analyze the set range of the score distribution of the abnormal score values ​​of each sample data in the initial sample dataset as the initial abnormal score threshold for the user; wherein the initial sample dataset is used as the historical sample dataset of the user's first health monitoring, and the initial abnormal score threshold is used as the abnormal score threshold of the user's first health monitoring.

[0168] In some embodiments, the model training module 305 is further configured to:

[0169] The current sample dataset is added to the historical sample dataset to serve as the historical sample dataset for the user's next health monitoring.

[0170] Based on the first algorithm and the second algorithm, the abnormal score value of each sample data in the historical sample dataset of the user's next health monitoring is obtained, and the abnormal score threshold of the user's next health monitoring is updated.

[0171] In some embodiments, the acquisition module 301 is further configured to acquire the user's feature data, which includes, but is not limited to, at least one of the following: first attribute data on whether the user is taking medication, second attribute data on whether the user suffers from a target disease, and third attribute data on the measurement method of the target physiological indicator; and to determine the data dimension of the user's sample data based on the feature data.

[0172] In practical applications, the acquisition module 301, sample construction module 302, scoring module 303, alarm module 304, and model training module 305 can be implemented by a processor in an electronic device. Of course, the processor needs to run a computer program in memory to implement its functions.

[0173] It should be noted that the above-described health monitoring-based alarm device is only illustrated by the division of the above-described program modules when performing health monitoring-based alarms. In practical applications, the above processing can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program modules to complete all or part of the processing described above. Furthermore, the health monitoring-based alarm device and the health monitoring-based alarm method embodiments provided above belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0174] Based on the hardware implementation of the above program modules, and in order to implement the method of the embodiments of this application, the embodiments of this application also provide an electronic device. Figure 4 The diagram shows only an exemplary structure of the electronic device, not the entire structure; implementation is possible as needed. Figure 4 The structure shown may be part or all of the structure.

[0175] like Figure 4 As shown, the electronic device 400 provided in this application embodiment includes: at least one processor 401, a memory 402, a user interface 403, and at least one network interface 404. The various components in the electronic device 400 are coupled together via a bus system 405. It can be understood that the bus system 405 is used to implement communication between these components. In addition to a data bus, the bus system 405 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 4 The general designated all buses as Bus System 405.

[0176] The user interface 403 may include a monitor, keyboard, mouse, trackball, click wheel, buttons, touchpad, or touch screen.

[0177] The memory 402 in this embodiment is used to store various types of data to support the operation of the electronic device. Examples of such data include any computer program used to operate on the electronic device.

[0178] The alarm method based on health monitoring disclosed in this application can be applied to, or implemented by, processor 401. Processor 401 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the alarm method based on health monitoring can be completed by the integrated logic circuitry in the hardware of processor 401 or by instructions in software form. The processor 401 can be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 401 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiments of this application can be directly manifested as execution by a hardware decoding processor, or execution by a combination of hardware and software modules in the decoding processor. The software modules can be located in a storage medium, specifically memory 402. Processor 401 reads information from memory 402 and, in conjunction with its hardware, completes the steps of the alarm method based on health monitoring provided in the embodiments of this application.

[0179] In an exemplary embodiment, the electronic device may be implemented by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers (MCUs), microprocessors, or other electronic components to perform the aforementioned method.

[0180] It is understood that memory 402 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), ferromagnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM); magnetic surface memory can be disk storage or magnetic tape storage. Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), SyncLink Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of this application are intended to include, but are not limited to, these and any other suitable types of memories.

[0181] In an exemplary embodiment, this application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a memory 402 that stores a computer program. The computer program can be executed by a processor 401 of an electronic device to complete the steps described in the method of this application embodiment. The computer-readable storage medium can be a ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM, etc.

[0182] It should be noted that terms such as "first" and "second" are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0183] Furthermore, the technical solutions described in the embodiments of this application can be combined arbitrarily without conflict.

[0184] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An alarm method based on health monitoring, characterized in that, include: Acquire at least one monitoring data point representing a user's physiological indicators; The current sample data is constructed based on at least one of the monitoring data; Based on the current sample data and historical sample dataset, a robust random segmentation forest algorithm is used to determine the anomaly score value of the current sample data; If the abnormal score value is determined to be outside the range of the abnormal score threshold, an alarm message is generated. Wherein, the number of data dimensions of each sample data in the current sample data and the historical sample dataset is d, where d is a natural number greater than or equal to 1, and the anomaly scoring threshold is determined based on the anomaly scoring value of each sample data in the historical sample dataset. The step of determining the anomaly score value of the current sample data using a robust random segmentation forest algorithm based on the current sample data and historical sample dataset includes: A first binary tree is constructed based on the first algorithm for the historical sample dataset; A second binary tree is constructed based on the second algorithm for the merged sample dataset of the historical sample dataset and the current sample data; For each sample data in the historical sample dataset, the sum of the depth differences between the second binary tree and the first binary tree for each sample data is calculated to obtain the anomaly score value of the current sample data; The step of constructing a second binary tree based on the merged sample dataset of the historical sample dataset and the current sample data using the second algorithm includes: The maximum, minimum, and range of the merged sample dataset in each data dimension are calculated. Determine a baseline value, wherein the value of the baseline value is greater than zero and less than the sum of the ranges of each data dimension; A second segmentation threshold is determined for the second target dimension used for binary tree partitioning. The second target dimension is the data dimension with the smallest index among d data dimensions whose cumulative range sum is greater than or equal to the benchmark value. The cumulative range sum is the sum of the ranges of the second target dimension and all data dimensions before the second target dimension, counted in the order of the data dimensions. The second segmentation threshold is the minimum value of the second target dimension plus the cumulative range sum minus the benchmark value. Compare the second segmentation threshold with the maximum and minimum values ​​of the second target dimension statistically obtained from the historical sample dataset in the first algorithm; If the second segmentation threshold is greater than or equal to the maximum value of the second target dimension of the historical sample dataset in the first algorithm, or less than or equal to the minimum value of the second target dimension of the historical sample dataset in the first algorithm, then a second binary tree is generated based on the comparison result of the value of the second target dimension of the current sample data with the second segmentation threshold and the first binary tree. The second binary tree includes the current sample data and the first binary tree located at the next level below the root node. If the value of the second target dimension of the current sample data is greater than the second segmentation threshold, then the current sample data is located on the right side of the first binary tree. If the value of the second target dimension of the current sample data is less than or equal to the second segmentation threshold, then the current sample data is located on the left side of the first binary tree. If the second segmentation threshold is less than the maximum value of the second target dimension of the historical sample dataset as counted in the first algorithm and greater than the minimum value of the second target dimension of the historical sample dataset as counted in the first algorithm, then the merged sample dataset, which serves as the root node, is partitioned into a binary tree based on the first algorithm to obtain the second binary tree.

2. The method according to claim 1, characterized in that, The construction of the first binary tree on the historical sample dataset based on the first algorithm includes: Calculate the maximum and minimum values ​​for each data dimension of the historical sample dataset; A first segmentation threshold is determined for a first target dimension used for binary tree partitioning, wherein the first target dimension is any one of d data dimensions, and the first segmentation threshold is less than the maximum value of the first target dimension and greater than the minimum value of the first target dimension. The historical sample dataset, which serves as the root node, is divided into a binary tree based on the first segmentation threshold. The first segmentation threshold of each node after the division is redefined. The binary tree is then divided into two parts based on the corresponding first segmentation threshold until the division is complete, thus obtaining the first binary tree.

3. The method according to claim 1, characterized in that, The method further includes: Construct the initial sample dataset for the user; The anomaly score value of each sample data in the initial sample dataset is obtained based on the first algorithm and the second algorithm; The distribution range of abnormal rating values ​​of each sample data in the initial sample dataset is statistically analyzed and used as the initial abnormal rating threshold for the user. Wherein, the initial sample dataset serves as the historical sample dataset for the user's first health monitoring, and the initial abnormal score threshold serves as the abnormal score threshold for the user's first health monitoring.

4. The method according to claim 1, characterized in that, After determining the anomaly score value of the current sample data using a robust random segmentation forest algorithm based on the current sample data and historical sample dataset, the method further includes: The current sample dataset is added to the historical sample dataset to serve as the historical sample dataset for the user's next health monitoring. Based on the first algorithm and the second algorithm, the abnormal score value of each sample data in the historical sample dataset of the user's next health monitoring is obtained, and the abnormal score threshold of the user's next health monitoring is updated.

5. The method according to claim 1, characterized in that, The method further includes: The user's characteristic data is obtained, which includes at least one of the following: first attribute data on whether the user is taking medication, second attribute data on whether the user suffers from the target disease, and third attribute data on the measurement method of the target physiological indicator. The data dimensions of the user's sample data are determined based on the feature data.

6. An alarm device based on health monitoring, characterized in that, include: The acquisition module is used to acquire at least one monitoring data point representing the user's physiological indicators; The sample construction module is used to construct the current sample data based on the at least one monitoring data. The scoring module is used to determine the outlier scores of the current sample data based on the current sample data and the historical sample dataset using a robust random segmentation forest algorithm. An alarm module is used to generate an alarm message if the abnormal score value is outside the numerical range of the abnormal score threshold. Wherein, the number of data dimensions of each sample data in the current sample data and the historical sample dataset is d, where d is a natural number greater than or equal to 1, and the anomaly scoring threshold is determined based on the anomaly scoring value of each sample data in the historical sample dataset. The scoring module is specifically used for: A first binary tree is constructed based on the first algorithm for the historical sample dataset; A second binary tree is constructed based on the second algorithm for the merged sample dataset of the historical sample dataset and the current sample data; For each sample data in the historical sample dataset, the sum of the depth differences between the second binary tree and the first binary tree for each sample data is calculated to obtain the anomaly score value of the current sample data; The construction of a second binary tree based on the merged sample dataset of the historical sample dataset and the current sample data using the second algorithm includes: The maximum, minimum, and range of the merged sample dataset in each data dimension are calculated. Determine a baseline value, wherein the value of the baseline value is greater than zero and less than the sum of the ranges of each data dimension; A second segmentation threshold is determined for the second target dimension used for binary tree partitioning. The second target dimension is the data dimension with the smallest index among d data dimensions whose cumulative range sum is greater than or equal to the benchmark value. The cumulative range sum is the sum of the ranges of the second target dimension and all data dimensions before the second target dimension, counted in the order of the data dimensions. The second segmentation threshold is the minimum value of the second target dimension plus the cumulative range sum minus the benchmark value. Compare the second segmentation threshold with the maximum and minimum values ​​of the second target dimension statistically obtained from the historical sample dataset in the first algorithm; If the second segmentation threshold is greater than or equal to the maximum value of the second target dimension of the historical sample dataset in the first algorithm, or less than or equal to the minimum value of the second target dimension of the historical sample dataset in the first algorithm, then a second binary tree is generated based on the comparison result of the value of the second target dimension of the current sample data with the second segmentation threshold and the first binary tree. The second binary tree includes the current sample data and the first binary tree located at the next level below the root node. If the value of the second target dimension of the current sample data is greater than the second segmentation threshold, then the current sample data is located on the right side of the first binary tree. If the value of the second target dimension of the current sample data is less than or equal to the second segmentation threshold, then the current sample data is located on the left side of the first binary tree. If the second segmentation threshold is less than the maximum value of the second target dimension of the historical sample dataset as counted in the first algorithm and greater than the minimum value of the second target dimension of the historical sample dataset as counted in the first algorithm, then the merged sample dataset, which serves as the root node, is partitioned into a binary tree based on the first algorithm to obtain the second binary tree.

7. An electronic device, characterized in that, include: A processor and memory for storing computer programs that can run on the processor, wherein, The processor, when running a computer program, performs the steps of the method according to any one of claims 1 to 5.

8. A computer storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images