Pension sign real-time monitoring method based on wearable device

Abstract

The invention relates to the technical field of intelligent sensors, in particular to a wearable device-based pension sign real-time monitoring method, which comprises the following steps of: acquiring sign data of an elderly user; determining a data point corresponding to each moment by using the sign data at the adjacent moments; determining the abnormal probability of the sign data based on the fitting curves of the sign data, and determining the abnormal possibility of the data points based on the abnormal probability and the difference between the trend terms of different fitting curves; the screened feature data points are clustered, and an abnormal cluster and a normal cluster are divided based on aggregation features of the feature data points in different clusters and intra-cluster distribution; and determining a correction result of the physical sign data based on the value conditions of the feature data points in the different clustering clusters, and obtaining a real-time monitoring result of the physical sign data. According to the method, the sign data is corrected by analyzing the sign data difference caused by the abnormal state of the elderly user and the movement of the wearable device, so that the accuracy of the monitoring data of the sign data is improved.

Description

Technical Field

[0001] This application relates to the technical field of intelligent sensors, and specifically to a real-time monitoring method for elderly physical signs based on wearable devices. Background Art

[0002] With the improvement of people's living standards, people pay more and more attention to their own health. It is particularly important to obtain the parameter indicators of the health status through some devices.

[0003] With the popularization of wearable devices, such as smart watches, bracelets, vests, etc., basic physiological indicators such as heart rate, blood pressure, and body temperature can be collected in real time through built-in biosensors. Wearable devices have good portability and simple operation, and have a high audience rate in the field of elderly health. Wearable devices combine data collection, transmission, and analysis through intelligent sensors to provide all-weather health protection for the elderly.

[0004] When using wearable devices to monitor the health status of the elderly in real time, after the physical sign data collected by the built-in sensors is analyzed and visualized, etc., it is convenient for the elderly to have a preliminary understanding of their own state, and it can also assist elderly care medical staff to conduct more professional monitoring and analysis. The physical sign monitoring data of wearable devices, such as heart rate, is generally obtained by collecting and calculating the blood data in the skin of the wearing part. However, due to factors such as possible displacement of the wearable device during the movement of the elderly or poor contact between the sensor and the skin, the collected data is not accurate, resulting in a certain deviation in the physical sign monitoring results. Summary of the Invention

[0005] In order to solve the problem that there is a certain deviation in the calculation result when calculating the heart rate of the human body by the existing method, the purpose of this application is to provide a real-time monitoring method for elderly physical signs based on wearable devices, and the specific technical solution adopted is as follows:

[0006] This application provides a real-time monitoring method for elderly physical signs based on wearable devices, and the method includes the following steps:

[0007] Collect physical sign data of elderly users by using wearable devices;

[0008] Determine the data points corresponding to each moment in the multi-dimensional data space by using the physical sign data collected at adjacent moments;

[0009] Determine the abnormal probability of each physical sign data at each moment based on the fitting curve of the physical sign data collected at each moment and historical moments, and determine the abnormal possibility of the data point corresponding to each moment based on the abnormal probability and the difference between the trend terms of the fitting curves of different physical sign data at each moment;

[0010] Cluster the characteristic data points screened for the abnormal possibility, and divide the abnormal clustering clusters and normal clustering clusters based on the aggregation characteristics of the characteristic data points within different clustering clusters in the clustering result and the distribution within the clusters;

[0011] Determine the correction result of the physical sign data at the corresponding moment of each characteristic data point within the abnormal clustering cluster based on the value of the characteristic data points within the abnormal clustering cluster and the normal clustering cluster, and obtain the real-time monitoring result of the physical sign data of the elderly user based on the correction result.

[0012] Preferably, the determining the data point corresponding to each moment in the multi-dimensional data space includes:

[0013] Take the ratio of the absolute value of the difference between each moment and the same physical sign data in the adjacent previous moment to the time interval between adjacent moments as the characteristic value of each moment in the same physical sign data;

[0014] Take each physical sign data as a dimension of the data space to construct a multi-dimensional data space; use the characteristic values of each moment in all physical sign data to determine a unique data point in the multi-dimensional data space.

[0015] Preferably, the determining the abnormal probability of each physical sign data at each moment includes:

[0016] Take the preset number of adjacent moments before each moment as adjacent historical moments, and take the curve fitting result of each physical sign data at each moment and the adjacent historical moments in time as the fitting curve of each physical sign data at each moment;

[0017] Take the product of the distribution variance of the fitting deviation of the fitting curve at each moment and the adjacent history and the sum of the fitting deviations as the abnormal probability of each physical sign data at each moment.

[0018] Preferably, the determining the abnormal possibility of the data point corresponding to each moment includes:

[0019] Use the STL method of time series decomposition to decompose the fitting curve of each physical sign data at each moment respectively to obtain the residual term and the trend term;

[0020] Calculate the difference distance between the trend terms of the fitting curves of any two different physical sign data at each moment respectively, and the difference distance between the residual terms, and perform a negative mapping on the difference distance between the residual terms;

[0021] Calculate the cumulative result of the product of the difference distance between the trend terms and the negative mapping result on all physical sign data at each moment as the monitoring abnormal degree of each moment;

[0022] Determine the anomaly possibility of the data points determined in the multi-dimensional data space at each moment by using the monitoring anomaly degree at each moment and the mean value of the anomaly probabilities of all types of physical sign data at each moment; wherein, the anomaly possibility has a negative correlation with the monitoring anomaly degree, and the anomaly possibility has a positive correlation with the mean value.

[0023] Preferably, the acquisition method of the characteristic data points is as follows: Obtain the threshold of the anomaly possibility of all data points in the multi-dimensional data space by using the global threshold segmentation method, and take the data points with the anomaly possibility greater than the threshold as the characteristic data points in the multi-dimensional data space.

[0024] Preferably, the division of the abnormal clustering clusters and the normal clustering clusters includes:

[0025] Determine the distribution anomaly index of each clustering cluster based on the difference in the concentration of the characteristic data points within each clustering cluster and the other clustering clusters and the distance between the cluster centers of the clustering clusters.

[0026] Calculate the distribution variance of the values of the characteristic data points within each clustering cluster in each dimension of the multi-dimensional data space, and take the product of the cumulative result of the distribution variance in all dimensions of the multi-dimensional data space and the distribution anomaly index as the difference index of each clustering cluster.

[0027] Obtain the segmentation threshold of the difference indexes of all clustering clusters by using the global threshold segmentation method, and take the clustering clusters with the difference index greater than the segmentation threshold as the abnormal clustering clusters; take the clustering clusters with the difference index less than or equal to the segmentation threshold as the normal clustering clusters.

[0028] Preferably, the determination of the distribution anomaly index of each clustering cluster includes:

[0029] Calculate the time interval between the corresponding acquisition moments of any two characteristic data points within each clustering cluster, and take the sum of the distribution variance of all the time intervals corresponding to each clustering cluster and the range between the characteristic data points within each clustering cluster as the within-class difference of each clustering cluster.

[0030] Calculate the cumulative sum of the Euclidean distances between the cluster centers of each clustering cluster and all the other clustering clusters in the multi-dimensional data space, and take the product of the normalized result of the cumulative sum of the Euclidean distances and the within-class difference as the distribution anomaly index of each clustering cluster.

[0031] Preferably, the determination of the correction result of the physical sign data at the moment corresponding to each characteristic data point within the abnormal clustering cluster includes:

[0032] Calculate the difference between the mean values of the values taken in the same dimension of the multi-dimensional data space of each normal clustering cluster and each abnormal clustering cluster as the numerator, and take the ratio of the numerator to the mean value of the values taken in the same dimension of the multi-dimensional data space of each abnormal clustering cluster as the first difference.

[0033] Take the absolute value of the difference between the distribution variances of the values of each normal clustering cluster and each abnormal clustering cluster on the same dimension in the multi-dimensional data space, and use the ratio of the absolute value of the difference to the distribution variance of the values of each abnormal clustering cluster on the same dimension in the multi-dimensional data space as the second difference;

[0034] Take the product of the first difference and the second difference as the correction coefficient of each normal clustering cluster to each abnormal clustering cluster in one dimension;

[0035] Take the mean value of the correction coefficients of all normal clustering clusters to each abnormal clustering cluster in all dimensions as the correction degree of each abnormal clustering cluster.

[0036] Preferably, obtaining the real-time monitoring result of the physical sign data of the elderly user based on the correction result includes:

[0037] For each feature data point in each abnormal clustering cluster, take the product of each physical sign data collected at the corresponding moment of each feature point and the correction degree of each abnormal clustering cluster as the correction amount of each physical sign data at the corresponding moment, and take the sum of the correction amount of each physical sign data and each physical sign data collected at the corresponding moment as the correction result of each physical sign data at the corresponding moment.

[0038] Preferably, obtaining the real-time monitoring result of the physical sign data of the elderly user based on the correction result includes:

[0039] Correct each physical sign data collected at the corresponding moment of each feature data point in the abnormal clustering cluster respectively;

[0040] For any kind of physical sign data, arrange all the correction results of each physical sign data and the data of each physical sign collected normally in chronological order to obtain the monitoring result of each physical sign data of the elderly user during the monitoring period.

[0041] This application has at least the following beneficial effects:

[0042] When considering the monitoring of human vital signs data by existing wearable devices, most of them use wearable devices to sample the vital signs data. This detection method is relatively convenient and thus widely used. However, since the wearable device may move during human movement, the collected vital signs data may differ from the actual change situation. Therefore, this application first determines the data points corresponding to each moment in the multi-dimensional data space by using the vital signs data collected at adjacent moments, and maps the vital signs data collected at each moment to the preliminary data space through the instantaneous change of the vital signs data at each moment, which is convenient for subsequent analysis from the data change rather than the collected data itself, and avoids the influence of isolated noise points. Secondly, based on the fitting curve of the vital signs data collected at each moment and historical moments, the abnormal probability of each type of vital signs data at each moment is determined, and based on the abnormal probability and the difference between the trend terms of the fitting curves of different types of vital signs data at each moment, the abnormal possibility of the data point corresponding to each moment is determined. By analyzing the different data changes caused by the movement of the wearable device and the abnormal situation of the elderly user himself, the abnormal probability of the vital signs data collected at each moment is evaluated. After that, the characteristic data points screened by the abnormal possibility are clustered, and the abnormal clustering clusters and normal clustering clusters are divided based on the aggregation characteristics of the characteristic data points in different clustering clusters and the distribution within the clusters in the clustering result. By clustering and dividing the characteristic data points, it is convenient for subsequent adaptive correction of the vital signs data at the moments corresponding to the characteristic data points in the abnormal clustering cluster by comparing the distribution differences between the characteristic data points in the normal clustering cluster and the abnormal clustering cluster. Finally, based on the value situations of the characteristic data points in the abnormal clustering cluster and the normal clustering cluster, the correction result of the vital signs data at the moment corresponding to each characteristic data point in the abnormal clustering cluster is determined, and based on the correction result, the real-time monitoring result of the elderly user's vital signs data is obtained. This application analyzes the overall and local change characteristics of the collected data, and adaptively corrects the abnormal vital signs data, thereby improving the accuracy of the vital signs data monitoring result. Brief Description of the Drawings

[0043] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following described drawings are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0044] Figure 1 It is a flowchart of the real-time monitoring method for elderly vital signs based on a wearable device provided by an embodiment of the present application. Detailed Embodiments

[0045] To further elaborate on the technical means and effects adopted by this application to achieve the intended invention purpose, the following provides a detailed description of the real-time monitoring method for elderly physical signs based on wearable devices proposed according to this application in combination with the accompanying drawings and preferred embodiments as follows.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which this application belongs.

[0047] The following specifically describes the specific solution of the real-time monitoring method for elderly physical signs based on wearable devices provided by this application in combination with the accompanying drawings.

[0048] Embodiment of the real-time monitoring method for elderly physical signs based on wearable devices:

[0049] The specific scenario targeted by this embodiment is as follows: When using wearable devices to monitor the health of elderly users, first, the wearable devices are worn on the elderly users. During the process of the wearable devices collecting the physical sign data of the elderly users through the built-in intelligent sensors, due to factors such as displacement and sweat of the wearable devices during the movement of the elderly users, which affect the accuracy of the collected data, resulting in deviations in the physical sign monitoring results. Therefore, it is necessary to analyze the overall characteristics and local characteristics of the collected data, and correct the abnormal data to a certain extent, so as to improve the accuracy of the physical sign monitoring results. Among them, the wearable devices include smart watches, smart bracelets, smart vests, etc.

[0050] Please refer to Figure 1 , which shows the flowchart of the real-time monitoring method for elderly physical signs based on wearable devices provided by an embodiment of this application. The method includes the following steps:

[0051] Step S1, use wearable devices to collect the physical sign monitoring data of elderly users.

[0052] This application aims to realize the physical sign monitoring of elderly users through wearable devices. First, the wearable devices are worn on the objects to be monitored to collect the physical sign data of elderly users, including but not limited to heart rate, blood pressure, body temperature, respiratory rate, and number of steps. In this embodiment, the preset collection duration is 10 minutes, and the data collection frequency is 60 Hz per minute, that is, the physical sign data is collected once per second. In other embodiments, the implementer can preset the collection duration and collection frequency according to the monitored physical signs of elderly users, and this application does not make special restrictions on this.

[0053] Step S2, analyze the difference between the changes in physical sign data caused by the movement of the wearable device and the changes in physical sign data caused by the abnormal state of the elderly user himself, calculate the abnormal possibility of the data points in the constructed multi-dimensional data space, and then determine the characteristic data points in the multi-dimensional data space.

[0054] When an elderly user wears a wearable device during exercise or when there is sweat between the wearable device and the skin, it will affect the accuracy of the collected physiological sign data. In this embodiment, the heart rate is taken as an example for analysis, the errors existing in the process of collecting heart rate data are analyzed, and the abnormal heart rate data is corrected to obtain accurate heart rate data.

[0055] Specifically, although the movement of the wearable device seems slight to humans, since skin textures and the like are at a fine level, once there is a slight movement, it may cause a change in the optical path of the sensor, resulting in a certain error in the collected data. Because during the heart rate monitoring process, the blood volume and the absorption change of hemoglobin in the human body are mainly collected, and then the heart rate result is calculated indirectly.

[0056] Further, for any moment during data collection, taking the i-th moment as an example for elaboration. First, the ratio of the absolute value of the difference between the number of steps collected at the i-th moment and the (i - 1)-th moment to the sampling interval is used as the speed eigenvalue at the i-th moment. Similarly, for the other types of physiological sign data collected by the wearable device, the ratio of the absolute value of the difference between the data collected at the i-th moment and the (i - 1)-th moment to the sampling interval is calculated respectively as the eigenvalue of the physiological sign data. Secondly, after obtaining the eigenvalues of all types of physiological sign data at each moment, the eigenvalues of each type of physiological sign data are mapped to one dimension to construct a multi-dimensional data space, and the eigenvalues of all types of physiological sign data at each moment determine a unique data point in the multi-dimensional data space.

[0057] Under normal circumstances, the contraction and relaxation of the heart show obvious periodicity, and the heart rate fluctuates within a certain range. When an elderly user is in a state of exercise, the heart rate, body temperature, and respiratory rate will also change significantly with the intensity of the exercise. Moreover, the heart rate, body temperature, and respiratory rate will increase with the intensity of the exercise and decrease with the smoothness of the exercise, that is, the collected data of different types of physiological signs show relatively similar trend characteristics within the same time period. However, when the distance between the wearable device and the skin of the elderly user changes during exercise, or when there is sweat at the wearing position, it will have different degrees of influence on different types of physiological sign data, resulting in the destruction of the above-mentioned similar trend characteristics.

[0058] Specifically, the first k moments of each moment are respectively taken as adjacent historical moments, and the collected data of each type of physiological sign at the first k moments and each moment are curve-fitted in time, and the fitting results are respectively used as the fitting curves of each type of physiological sign data at each moment.

[0059] Secondly, after obtaining the fitting curves of all types of vital sign data at each moment, the fitting curves of each type of vital sign data at each moment are decomposed respectively by using the STL method of time series decomposition to obtain the residual term, the trend term, and the seasonal term. Among them, the STL method is a well-known technology, and the specific process will not be elaborated here.

[0060] After that, the difference distances between the trend terms and between the residual terms of the fitting curves of different types of vital sign data at the i-th moment are calculated respectively. Among them, the difference distance is used to measure the difference degree between the trend terms and between the residual terms. The difference distances that can be used in different embodiments include but are not limited to the Euclidean distance, the DTW distance, and the value variance.

[0061] It should be noted that curve fitting is a common technique in the field of data processing, and the specific process will not be elaborated here. The common curve fitting methods include but are not limited to least squares fitting and polynomial fitting. This application does not make special limitations on the specific method of curve fitting. Preferably, in an embodiment of this application, the fitting curves of each type of vital sign data at k moments are obtained by using the least squares fitting method.

[0062] Furthermore, it should be noted that obtaining the data of k historical moments at each moment is to analyze the data change trend at each moment, and can avoid the influence of noise points on the data of a single moment, and improve the accuracy of subsequent data analysis. On the premise that the above purpose can be achieved, different values can be selected for the size of k in different embodiments, and this application does not make special restrictions on this. Preferably, in an embodiment of this application, the size of k is set to 7. For the moments with the number of historical moments less than k, the data of k historical moments are obtained by using data filling methods, and the data filling methods include but are not limited to nearest neighbor filling, mean filling, interpolation filling, etc. Data filling is a well-known technology, and the specific process will not be elaborated here.

[0063] Furthermore, when using a wearable device to monitor the vital signs of elderly users, the elderly are usually guided to perform some health-preserving actions, such as walking slowly and doing health-care exercises, etc., so as to more accurately reflect the health status of the elderly users. If there is a change in the distance between the wearable device and the skin of the elderly user during exercise, or if sweat is generated at the wearing site and affects different types of vital sign data, and the duration of the elderly user even in the exercise recovery stage is usually relatively long, so it should have an impact on the vital sign data at several consecutive acquisition moments, and with the change of the exercise intensity or the duration of exercise, the degree of influence on the vital sign data will also change accordingly.

[0064] First, obtain the fitting curve at the i-th moment, and calculate the fitting deviations at k + 1 moments on the fitting curve respectively. The greater the difference in the fitting deviations among the k + 1 moments, the more obvious the change in the instantaneous trend characteristics of the vital sign data at different moments; the smaller the fitting deviations at the k + 1 moments and the smaller the difference in the fitting deviations among the k + 1 moments, the smoother the trend of the vital sign data among the k + 1 moments. Here, the abnormal probability of the vital sign data at each moment is initially evaluated through the similarity of the trend characteristics between the fitting curves of different vital sign detection data at the same moment, and the abnormal probability of the data at each moment is obtained. Calculate the abnormal probability of the j-th type of vital sign data at the i-th moment:

[0065] t i,j = σ i,j × z i,j

[0066] In the formula, t i,j is the abnormal probability of the j-th type of vital sign data at the i-th moment, σ i,j is the distribution variance of the fitting deviations at k + 1 moments on the fitting curve of the j-th type of vital sign data at the i-th moment, and z i,j is the sum of the fitting deviations at k + 1 moments on the fitting curve of the j-th type of vital sign data at the i-th moment.

[0067] Secondly, combining the stability of the trend characteristics of the fitting curves of various vital sign data at adjacent moments, and the size of the time interval between the data points corresponding to each moment in the multi-dimensional data space, further evaluate the abnormal possibility of each data point. First, for the i-th moment, calculate the difference distance d i,1 between the trend terms of the fitting curves of any two different types of vital sign data at the i-th moment, and the difference distance d i,2 between the residual terms. Use the negative mapping result of d i,2 as the weight to weight d i,1 . The accumulated result of the weighted result on all types of vital sign data at the i-th moment is used as the monitoring abnormal degree at the i-th moment. The reason for such calculation is that the smaller the residual term, the less interference from factors such as noise during data acquisition, and the higher the data accuracy; the smaller the difference distance between the residual terms, the more consistent the interference degrees of the two types of vital sign data at the i-th moment.

[0068] After that, use the monitoring abnormal degree at the i-th moment and the mean value of the abnormal probabilities of all types of vital sign data at the i-th moment to determine the abnormal possibility of the data point p determined by the eigenvalues of all types of vital sign data at the i-th moment in the multi-dimensional data space. Among them, the abnormal possibility is negatively correlated with the monitoring abnormal degree and positively correlated with the mean value.

[0069] It should be noted that positive correlation means that when one variable increases, the other variable also increases, and the two variables change in the same direction. When one variable changes from large to small or from small to large, the other variable also changes from large to small or from small to large; the specific relationship is determined by the actual application, and this application does not make special restrictions.

[0070] It should be noted that negative correlation means that when one variable increases, the other variable decreases accordingly, and the two variables change in opposite directions. When one variable changes from large to small or from small to large, the other variable also changes from small to large or from large to small; the specific relationship is determined by the actual application, and this application does not make special restrictions.

[0071] Preferably, in an embodiment of this application, the formula for calculating the anomaly possibility of the data point p determined in the multi-dimensional data space at the i-th moment is as follows:

[0072]

[0073] In the formula, r i is the monitoring anomaly degree at the i-th moment, M is the type of physical sign data, j and n respectively represent the j-th and n-th physical sign detection data, d i,1 (j, n), d i,2 (j, n) are respectively the difference distance between the trend terms and the difference distance between the residual terms of the fitting curves of the j-th and n-th physical sign detection data at the i-th moment. μ is a constant parameter used to prevent the denominator from being 0, and it can take the value of 0.001 here. The specific value can be adjusted according to the implementation scenario, and this application does not make special restrictions;

[0074] u p is the anomaly possibility of the data point p, is the mean value of the anomaly probabilities of all types of physical sign data at the i-th moment.

[0075] Furthermore, according to the above method, the anomaly possibility of each data point in the multi-dimensional data space can be obtained. The greater the anomaly possibility, the higher the possibility that the data point is abnormal. The global threshold segmentation method is used to obtain the threshold of all the anomaly possibilities, and the data points with anomaly possibility greater than the threshold are used as feature data points. Among them, threshold segmentation is a commonly used technique in the field of data processing, and the specific process will not be elaborated here.

[0076] Step S3, perform clustering division on the feature data points in the multi-dimensional data space, and divide the abnormal clustering clusters and normal clustering clusters according to the aggregation characteristics of the feature data points within different clustering clusters and the distribution within the clusters.

[0077] Specifically, if the problem at the i-th moment is caused by a sudden abnormality in the elderly user's own health condition during the physical sign monitoring process, then the various characteristic data collected by the sensors in the wearable device will change significantly at the i-th moment. Moreover, during the entire physical sign monitoring cycle, the frequency of sudden abnormalities in the elderly user's own health condition should be very low, and the distribution of corresponding data points in the multi-dimensional data space is not concentrated. Regarding the change in the distance between the wearable device and the elderly user's skin during movement, or the influence of sweat generated at the wearing position on the collection of different physical sign data by the sensors, due to the intermittent characteristics of movement, the distribution of corresponding data points in the multi-dimensional data space has a certain degree of concentration. For example, if the elderly user takes a slow walk for 5 minutes after a meal, then the data points corresponding to the physical sign data that change due to movement within a few minutes after each of the three meals a day for the elderly user have obvious aggregation.

[0078] Further, after obtaining the characteristic data points in the multi-dimensional data space according to the above steps, use a clustering algorithm to divide all the characteristic data points into several clustering clusters.

[0079] It should be noted that data clustering is a common technique in the field of data processing, and the specific process will not be elaborated here. Common data clustering algorithms include but are not limited to K-means clustering, AP (Affinity Propagation) clustering, and hierarchical clustering. The application does not impose special restrictions on the clustering method of the characteristic data points. Preferably, the application uses the AP clustering algorithm to obtain the clustering result of the characteristic data points.

[0080] Secondly, for any one clustering cluster, take the a-th clustering cluster as an example for subsequent analysis. If the intervals between the moments corresponding to the characteristic data points within the a-th clustering cluster are basically close and have a relatively stable periodicity, then the characteristic data points corresponding within the a-th clustering cluster are more likely to be the different collection moments during the elderly user's regular daily exercise; if the differences in the intervals between the moments corresponding to the characteristic data points within the a-th clustering cluster are relatively large, and the discreteness of the data points within the a-th clustering cluster is relatively strong, then the characteristic data points within the a-th clustering cluster are more likely to be the moments when the elderly user's own health condition suddenly abnormalities during the physical sign monitoring process.

[0081] Here, calculate the difference index of each clustering cluster, which is used to reflect the difference in the concentration of the distribution of characteristic data points between each clustering cluster and the other clustering clusters. The calculation formula for the difference index of the a-th clustering cluster is:

[0082] L a =norm(d a )×(σ(T a )+C a )

[0083] In the formula, L ais the distribution anomaly index of the a-th cluster, d a is the sum of the Euclidean distances between the cluster center of the a-th cluster and the cluster centers of the remaining clusters in the multi-dimensional data space, and norm() is the normalization function, T a is the set composed of the time intervals between the acquisition times corresponding to the feature data points within the a-th cluster, σ(T a ) is the set T a The variance of the time intervals within, C a is the range between the feature data points within the a-th cluster;

[0084] Secondly, combined with the difference situation of the values of the feature data points within the a-th cluster in each dimension in the multi-dimensional data space, the difference index of the a-th cluster is determined:

[0085]

[0086] In the formula, V a is the difference index of the a-th cluster, M is the dimension of the multi-dimensional data space, which is equal to the number of types of the collected physical sign data, m is the m-th dimension of the multi-dimensional data space, and σ c,m is the distribution variance of the values of the feature data points within the a-th cluster in the m-th dimension.

[0087] Among them, (σ(T a ) + C a ) reflects the intra-class difference of the a-th cluster. The intra-class difference characterizes the periodicity of the time intervals corresponding to the feature data points within the a-th cluster and the degree of discreteness of the positions. Then, the distance d a between the a-th cluster and the remaining clusters is normalized, and the intra-class difference is weighted to obtain the distribution anomaly index L a . The larger the value, the more irregular the distribution of the feature data points within the cluster; and the value of the feature data point in the same dimension is the feature value of a type of physical sign data corresponding to the acquisition time. Therefore, the larger the value of the distribution variance σ c,m , the greater the difference in the physical sign data changes of the data points within the a-th cluster at adjacent times.

[0088] Furthermore, the difference index of each cluster is obtained respectively, and the segmentation threshold of the difference indexes of all clusters is obtained by using the global threshold segmentation method. The clusters with the difference index greater than the segmentation threshold are used as the abnormal clusters affected by the movement of the wearable device, and the clusters with the difference index less than or equal to the segmentation threshold are used as the normal clusters not affected by the movement of the wearable device. Among them, the global threshold segmentation is a commonly used technology in the field of data processing, and the specific process will not be elaborated here.

[0089] S4. Determine the correction result of the vital sign data at the corresponding moment of each feature data point in the abnormal clustering cluster based on the value conditions of the feature data points in the abnormal clustering cluster and the normal clustering cluster, so as to obtain the real-time monitoring result of the elderly user's vital sign data.

[0090] In the embodiment of the present application, an abnormal clustering cluster is screened out in step S3. The heart rate corresponding to the collection moment of the feature data points in the abnormal clustering cluster is more likely to be abnormal data affected by the movement of the wearable device. Therefore, in the embodiment of the present application, the heart rate at the collection moment corresponding to the feature data points in the abnormal clustering cluster will be corrected. For the abnormal clustering cluster, the greater the deviation from the normal vital sign data collected at the corresponding moment of the feature data points in the normal clustering cluster, the greater the degree of correction required.

[0091] Here, calculate the correction degree of each abnormal clustering cluster, which is used to characterize the degree of correction required compared to the vital sign data collected at the corresponding moment of the feature data points in the normal clustering cluster.

[0092] Specifically, for the b-th abnormal clustering cluster, the specific calculation formula for the correction degree corresponding to the b-th abnormal clustering cluster is:

[0093]

[0094] In the formula, q b,a (m) is the correction coefficient of the a-th normal clustering cluster to the b-th abnormal clustering cluster in the m-th dimension, are the means of the values of all feature data points in the a-th normal clustering cluster and the b-th abnormal clustering cluster in the m-th dimension respectively, and σ a,m , σ b,m are the distribution variances of the values of all feature data points in the a-th normal clustering cluster and the b-th abnormal clustering cluster in the m-th dimension respectively;

[0095] Q b is the correction degree of the b-th abnormal clustering cluster, and N is the number of normal clustering clusters.

[0096] It should be noted that the distribution variance reflects the degree of data fluctuation. The second difference characterizes the distribution deviation of the values of the feature data points in the b-th abnormal clustering cluster in the m-th dimension compared to the a-th normal clustering cluster, and the first difference reflects the numerical deviation of the values of the feature data points in the two clustering clusters in the m-th dimension. The greater the distribution deviation and data deviation, the more seriously the feature data points at the corresponding collection moment in the b-th abnormal clustering cluster are affected by the movement of the wearable device, and the higher the correction degree.

[0097] Further, according to the above process, the correction degree of each abnormal clustering cluster is obtained respectively, and the collected heart rate data is corrected based on the correction degree.

[0098] Specifically, for each feature data point in the b-th abnormal clustering cluster, the product of each type of vital sign data collected at the corresponding moment of each feature point and the correction degree of the b-th abnormal clustering cluster is used as the correction amount of each type of vital sign data at the corresponding moment, and the sum of the correction amount of each type of vital sign data and each type of vital sign data collected at the corresponding moment is used as the correction result of each type of vital sign data at the corresponding moment.

[0099] After that, for any one feature data point in each abnormal clustering cluster, the vital sign data collected at the corresponding moment of each feature data point in the abnormal clustering cluster is corrected respectively according to the above steps, so as to obtain the correction results of the vital sign data collected at the corresponding moment of the feature data points in all abnormal clustering clusters. Finally, for any type of vital sign data, in chronological order, all the correction results of each type of vital sign data are inserted into the normally collected data to obtain the monitoring results of each type of vital sign data of the elderly user during the monitoring period.

[0100] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

[0101] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. The key point of each embodiment is to illustrate the differences from other embodiments.

Claims

1. A real-time monitoring method for elderly vital signs based on wearable devices, characterized in that: The method comprises the following steps: Use wearable devices to collect vital sign data of elderly users; Using the vital sign data collected at adjacent moments, determine the data point corresponding to each moment in the multidimensional data space; Determine the abnormal probability of each type of vital sign data at each moment based on the fitting curve of the vital sign data collected at each moment and at historical moments, and determine the abnormal possibility of the corresponding data point at each moment based on the abnormal probability and the difference between the trend items of the fitting curves of different types of vital sign data at each moment; Clustering the characteristic data points selected for the abnormal possibility, and dividing the abnormal clustering clusters and normal clustering clusters based on the aggregation characteristics and intra-cluster distribution of the characteristic data points in different clustering clusters in the clustering results; Based on the values ​​of the characteristic data points in the abnormal clusters and the normal clusters, the correction results of the vital sign data of each characteristic data point in the abnormal clusters at the corresponding time are determined, and the real-time monitoring results of the vital sign data of the elderly user are obtained based on the correction results.

2. The real-time monitoring method for elderly vital signs based on wearable devices according to claim 1 is characterized in that: Determining the data point corresponding to each moment in the multidimensional data space includes: The ratio of the absolute value of the difference between the same type of vital sign data at each moment and the previous moment to the time interval between adjacent moments is taken as the characteristic value of the same type of vital sign data at each moment; Take each type of vital sign data as a dimension of the data space to construct a multidimensional data space; use the characteristic values ​​of all types of vital sign data at each moment to determine a unique data point in the multidimensional data space.

3. The real-time monitoring method for elderly vital signs based on wearable devices according to claim 1 is characterized in that: Determining the abnormal probability of each type of vital sign data at each moment includes: A preset number of adjacent moments before each moment are taken as adjacent historical moments, and a curve fitting result of each vital sign data at each moment and the adjacent historical moments in time is taken as a fitting curve of each vital sign data at each moment; The product of the distribution variance of the fitting deviation of the fitting curve at each moment and in the adjacent history and the cumulative sum of the fitting deviation is taken as the abnormal probability of each vital sign data at each moment.

4. The real-time monitoring method for elderly vital signs based on wearable devices according to claim 1 is characterized in that: Determining the abnormal possibility of the data point corresponding to each moment includes: The time series decomposition STL method is used to decompose the fitting curve of each type of vital sign data at each moment, and the residual term and trend term are obtained respectively; Calculating the difference distance between trend items and the difference distance between residual items of the fitting curves of any two different vital sign data at each moment respectively, and performing negative mapping on the difference distance between the residual items; Calculate the product of the difference distance between the trend items and the negative mapping result on all kinds of vital sign data at each moment as the monitoring abnormality degree at each moment; The abnormal probability of the data point determined at each moment in the multidimensional data space is determined by using the monitoring abnormality at each moment and the mean of the abnormal probability of all kinds of vital sign data at each moment; wherein the abnormal possibility is negatively correlated with the monitoring abnormality, and the abnormal possibility is positively correlated with the mean.

5. The real-time monitoring method for elderly vital signs based on wearable devices according to claim 1 is characterized in that: The characteristic data points are obtained by using a global threshold segmentation method to obtain a threshold value of the abnormal possibility of all data points in the multidimensional data space, and taking data points whose abnormal possibility is greater than the threshold value as characteristic data points in the multidimensional data space.

6. The real-time monitoring method for elderly vital signs based on wearable devices according to claim 1 is characterized in that: The dividing into abnormal clusters and normal clusters includes: Determine the distribution anomaly index of each cluster based on the difference in the distribution concentration of feature data points in each cluster and other clusters and the distance between the center points of the clusters; Calculate the distribution variance of the characteristic data points in each cluster in each dimension of the multidimensional data space, and use the product of the cumulative result of the distribution variance in all dimensions of the multidimensional data space and the distribution anomaly index as the difference index of each cluster; The segmentation threshold of the difference index of all clusters is obtained by global threshold segmentation. The clusters with difference index greater than the segmentation threshold are regarded as abnormal clusters; the clusters with difference index less than or equal to the segmentation threshold are regarded as normal clusters.

7. The real-time monitoring method for elderly vital signs based on wearable devices according to claim 6 is characterized in that: The determining of the distribution anomaly index of each cluster includes: Calculate the time interval between the corresponding collection moments of any two characteristic data points in each cluster, and take the sum of the distribution variance of all the time intervals corresponding to each cluster and the range between the characteristic data points in each cluster as the intra-class difference of each cluster; The cumulative sum of the Euclidean distances between each cluster and the cluster centers of all other clusters in the multidimensional data space is calculated, and the product of the normalized result of the cumulative sum of the Euclidean distances and the intra-class difference is used as the distribution anomaly indicator of each cluster.

8. The real-time monitoring method for elderly vital signs based on wearable devices according to claim 1 is characterized in that: The step of determining the correction result of the vital sign data at the corresponding time of each characteristic data point in the abnormal cluster includes: Calculate the difference between the mean values ​​of each normal cluster and each abnormal cluster in the same dimension of the multidimensional data space as the numerator, and take the ratio of the numerator to the mean value of each abnormal cluster in the same dimension of the multidimensional data space as the first difference; The absolute value of the difference between the distribution variance of each normal cluster and each abnormal cluster in the same dimension of the multidimensional data space is used, and the ratio of the absolute value of the difference to the distribution variance of each abnormal cluster in the same dimension of the multidimensional data space is used as the second difference; The product of the first difference and the second difference is used as a correction coefficient of each normal cluster to each abnormal cluster in one dimension; The mean of the correction coefficients of all normal clusters for each abnormal cluster in all dimensions is taken as the correction degree of each abnormal cluster.

9. The real-time monitoring method for elderly vital signs based on wearable devices according to claim 1 is characterized in that: The real-time monitoring result of the elderly user's vital sign data obtained based on the correction result includes: For each characteristic data point in each abnormal cluster, the product of each type of vital sign data collected at the corresponding moment of each characteristic point and the correction degree of each abnormal cluster is used as the correction amount of each vital sign data at the corresponding moment, and the sum of the correction amount of each vital sign data and each vital sign data collected at the corresponding moment is used as the correction result of each vital sign data at the corresponding moment.

10. The real-time monitoring method for elderly vital signs based on wearable devices according to claim 1 is characterized in that: The real-time monitoring result of the elderly user's vital sign data obtained based on the correction result includes: Correct each type of vital sign data collected at the corresponding time of each characteristic data point in the abnormal cluster respectively; For any kind of vital sign data, all correction results of each kind of vital sign data and the normally collected data of each kind of vital sign are arranged in chronological order to obtain the monitoring results of each kind of vital sign data of the elderly user during the monitoring period.

Images