Pension health data analysis method and system based on big data and storage medium

Abstract

The invention relates to the technical field of big data processing, and discloses an old-age care health data analysis method and system based on big data and a storage medium. The method comprises the steps of monitoring a tracking target based on a first target scheme, and obtaining body data and behavior data; performing correlation analysis on the body data and the behavior data, and dividing the behavior data into positive influence data and negative influence data; generating a suggestion scheme based on the influence weights of the positive influence data and the negative influence data and the occurrence frequency in the behavior data; generating a correction strategy based on the suggested scheme, reselecting a monitoring scheme as a second target scheme, and continuously obtaining actual behavior data of the tracking target based on the second target scheme; and analyzing the actual behavior data to predict a future behavior mode of the tracking target, and correcting the future behavior mode in combination with a correction strategy. According to the invention, in combination with the body data and the historical behavior data of the elderly, a personalized health suggestion scheme can be given.

Description

Technical Field

[0001] The present application relates to the field of big data processing technology, and in particular to a method, system and storage medium for analyzing elderly health data based on big data. Background Art

[0002] Currently, community-based elderly care has initially achieved the collection and management of the living conditions and health information of the elderly through the introduction of smart devices, health monitoring systems and community management platforms. However, these technologies are mostly single-function-based, usually staying at the level of basic health data collection and offline services, making it difficult to fully analyze and efficiently meet the personalized needs of the elderly.

[0003] As an important part of modern information technology, big data technology has been initially applied in the field of community elderly care. By integrating and analyzing the health data, living habits data and social interaction data of the elderly, real-time monitoring and dynamic evaluation of the health status of the elderly can be achieved. For example, the Chinese patent document with publication number CN117457227A discloses an intelligent analysis method, system, equipment and medium for health data of elderly people. The method constructs a health data management model for different disease types; by obtaining the initial health data of the elderly in the specified collection period, and according to the health data management model of different disease types, the high-risk disease type of the current elderly is obtained, and then the corresponding health management plan for the current elderly is obtained; finally, by obtaining the behavior tracking data of the current elderly, the current elderly can be warned of dangerous behavior status. For example, a Chinese patent document with publication number CN117116480A discloses a health data monitoring and analysis system for elderly care workers. The system screens out the age range with the largest proportion of elderly people with the lowest health level of each health indicator, and provides targeted health guidance to the elderly in this age range. At the same time, the system performs comprehensive health coefficient statistics based on the mean value of different health indicators of each elderly person. The statistical result value realizes the quantification of the health status of the elderly, which is convenient for relevant staff to intuitively understand the comprehensive physical condition of each elderly person.

[0004] However, the health baselines and behavioral habits of different elderly people are different. The general models constructed by the above two methods may not be able to adapt to individual differences, making it difficult to provide personalized health advice plans. Summary of the invention

[0005] In order to fully and accurately serve the health management needs of the elderly, this application provides a big data-based elderly health data analysis method, system and storage medium.

[0006] In order to achieve the above-mentioned purpose of the invention, the present invention proposes a method for analyzing elderly health data based on big data, including:

[0007] Setting multiple monitoring schemes, obtaining basic data of the tracking target, and selecting one of the monitoring schemes as the first target scheme based on the basic data;

[0008] Monitoring the tracking target based on the first target plan to obtain physical data and behavioral data of the tracking target in a predetermined time period;

[0009] Performing correlation analysis on the body data and the behavior data, and dividing the behavior data affecting the body data into positive impact data and negative impact data;

[0010] The positive impact data and the negative impact data have corresponding impact weights, and a suggestion scheme is generated based on the impact weights of the positive impact data and the negative impact data and the frequencies of occurrence in the behavior data;

[0011] Generate a correction strategy based on the suggested solution, reselect one of the monitoring solutions as a second target solution, and continuously obtain actual behavior data of the tracking target based on the second target solution;

[0012] The actual behavior data is analyzed to predict the future behavior pattern of the tracking target, and the future behavior pattern is corrected in combination with the correction strategy.

[0013] Further, selecting the second target solution includes the following steps:

[0014] Extracting dynamic statistical features of the body data based on a sliding window method, numerically mapping the dynamic statistical features based on a fuzzy algorithm to obtain a fuzzy membership, mapping the fuzzy membership to rules based on expert knowledge rules to obtain a rule score, calculating the average of all rule scores of the same type, and weighted summing the averages of rule scores of different types based on the medical history data in the basic data to obtain a comprehensive status score, and performing threshold division on the comprehensive status score to divide the tracking target into low risk, medium risk and high risk;

[0015] The monitoring schemes are divided into low-frequency schemes, medium-frequency schemes and high-frequency schemes; the corresponding monitoring schemes are selected as alternative schemes based on the risk classification results of the tracking target; a user feature matrix is ​​constructed based on the physical data and the behavioral data of the tracking target; a scheme feature matrix is ​​constructed based on the monitoring frequency, monitoring time period and monitoring type in the alternative schemes; the adaptation scores of the alternative schemes are calculated based on the user feature matrix and the scheme feature matrix; and the alternative scheme with the highest adaptation score is used as the second target scheme.

[0016] Furthermore, performing association analysis includes the following steps:

[0017] The missing values ​​of the physical data are supplemented based on the linear interpolation method to obtain the first data, the behavioral data are encoded to obtain the second data, representative features in the first data are extracted, interactive features are generated based on the representative features and the second data, fusion features are generated by splicing the interactive features and the second data, and the fusion features are analyzed based on linear regression to obtain positive impact data and negative impact data and the corresponding impact weights.

[0018] Furthermore, correcting the future behavior pattern comprises the following steps:

[0019] Acquire historical data, wherein the historical data includes the behavioral data and environmental data of past dates, generate multiple predicted behavior patterns for future dates based on the historical data and the actual behavior data, select the predicted behavior pattern with the highest probability of occurrence as the future behavior pattern, the correction strategy includes multiple corrective behaviors, and when there is a predicted behavior in the future behavior pattern that violates the corrective behavior, generate a warning reminder to complete the correction of the future behavior pattern.

[0020] Further, generating the predicted behavior pattern comprises the following steps:

[0021] Setting date attributes, the date attributes including working days, rest days and holidays, setting date combination rules, establishing a prediction model based on the historical data, defining the dates two days before the future date as the first date and the second date respectively, obtaining the date attributes of the first date and the second date and defining them as the first attribute and the second attribute respectively, and determining the third date according to the first attribute, the second attribute and the date combination rule;

[0022] Compare the historical data of the second date and the third date. If the similarity between the two is less than the first threshold, determine the fourth date again based on the date combination rule, continue to compare the historical data with the same date attributes, locate and eliminate the abnormal data, and then input the remaining historical data into the prediction model to obtain the predicted behavior pattern.

[0023] Furthermore, comparing the historical data comprises the following steps:

[0024] The two pieces of historical data to be compared are respectively converted into a first sequence and a second sequence, and the same behaviors of the first sequence and the second sequence are compared. If the number of the same behaviors is less than a second threshold, the similarity between the first sequence and the second sequence is set to 0; otherwise, the similarity between the first sequence and the second sequence is calculated based on the overlapping time length of the same behaviors.

[0025] Furthermore, the unselected predicted behavior pattern is used as an alternative pattern, and a behavior sequence is generated based on the real-time behavior data of the future date. If the mutual exclusivity between the behavior sequence and the future behavior pattern is greater than a second threshold, one of the alternative patterns is reselected as the future behavior pattern.

[0026] Furthermore, the prediction model includes a local model and a cloud model, the local model extracts comprehensive features of the body data and the behavior data, encrypts the comprehensive features and uploads them to the cloud model, and the cloud model generates the predicted behavior pattern based on the comprehensive features.

[0027] The present application provides a big data-based elderly care health data analysis system, which is used to implement the above-mentioned big data-based elderly care health data analysis method, and the system includes:

[0028] A tracking module, which is provided with a plurality of monitoring schemes, obtains basic data of a tracking target, selects one of the monitoring schemes as a first target scheme based on the basic data, monitors the tracking target based on the first target scheme, and obtains physical data and behavioral data of the tracking target in a predetermined time period;

[0029] an analysis module, performing correlation analysis on the body data and the behavior data, dividing the behavior data affecting the body data into positive impact data and negative impact data, wherein the positive impact data and the negative impact data have corresponding impact weights, and generating a suggestion plan based on the impact weights of the positive impact data and the negative impact data and the occurrence frequency in the behavior data;

[0030] A correction module generates a correction strategy based on the suggested solution, reselects one of the monitoring solutions as a second target solution, and continuously obtains actual behavior data of the tracking target based on the second target solution;

[0031] The correction module analyzes the actual behavior data to predict the future behavior pattern of the tracking target, and corrects the future behavior pattern in combination with the correction strategy.

[0032] The present application provides a computer-readable storage medium having instructions stored thereon, and when the instructions are executed by a processor, the method for analyzing elderly health data based on big data as described above is implemented.

[0033] By setting up multiple monitoring schemes and obtaining the basic data of the tracking target, the present invention can select the optimal monitoring scheme as the first target scheme according to the personalized health status of the tracking target, thereby realizing health monitoring with stronger pertinence and adaptability. According to the reasonable configuration of monitoring functions and monitoring frequencies, different monitoring needs can be flexibly responded to. On the basis of obtaining physical data and behavioral data, the behavioral data is divided into positive impact data and negative impact data through association analysis, and the corresponding impact weights are assigned to them. Finally, a targeted optimization suggestion scheme can be generated based on the frequency of occurrence of behavioral data.

[0034] The present invention generates a correction strategy based on the suggested solution, and dynamically adjusts the monitoring solution to the second target solution, so as to continuously optimize the monitoring method based on real-time data tracking. By analyzing actual behavior data to predict future behavior patterns, and combining correction strategies to intervene in bad behaviors in advance, a closed-loop management from data collection to behavior intervention is achieved, which greatly improves the effectiveness and intelligence level of health management. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0036] Figure 1 This is a schematic diagram of the big data-based elderly care health data analysis method for this application;

[0037] Figure 2 This is a structural diagram of the big data-based elderly care health data analysis system for this application. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0039] It is understood that the terms "first", "second", etc. used in this application may be used herein to describe various elements, but unless otherwise specified, these elements are not limited by these terms. These terms are only used to distinguish a first element from another element. For example, without departing from the scope of this application, a first xx script may be referred to as a second xx script, and similarly, a second xx script may be referred to as a first xx script.

[0040] like Figure 1 As shown, a method for analyzing elderly care health data based on big data includes:

[0041] S1: Set up multiple monitoring schemes, obtain basic data of the tracking target, and select a monitoring scheme as the first target scheme based on the basic data.

[0042] The monitoring scheme includes monitoring functions and the monitoring frequency of each monitoring function. The monitoring functions include heart rate, blood pressure, blood oxygen, diet and medication use. Different monitoring frequencies correspond to certain function settings. For example, the heart rate can be detected every half an hour or every ten minutes. Some monitoring schemes do not have the function of monitoring medication use. The tracking target is the elderly person to be tracked. The basic data of the tracking target includes medical history, historical blood sugar and blood pressure conditions, activity preferences, etc. If the tracking target has diabetes and other conditions, a monitoring scheme including an image recognition function is used. A camera is installed in the tracking target's home, and the diet intake category of the tracking target is obtained through image recognition, thereby determining its sugar intake. The present invention uses the monitoring scheme selected based on the basic data as the first target scheme.

[0043] S2: monitoring the tracking target based on the first target plan, and obtaining the physical data and behavioral data of the tracking target in a predetermined time period.

[0044] S3: Perform correlation analysis on physical data and behavioral data, and divide the behavioral data that affect physical data into positive impact data and negative impact data.

[0045] The scheduled time period is one month, and the corresponding physical data and behavioral data are collected according to the collection frequency and function of the first target plan. Among them, heart rate, blood pressure, blood oxygen, body temperature, sleep quality, etc. belong to physical data, and number of steps, exercise duration, exercise content, movement trajectory, diet content, drug use records, etc. belong to behavioral data. The above data can be collected through various IoT devices, such as smart watches, smart cameras, smart medicine boxes, etc.

[0046] After that, the above data is subjected to correlation analysis. According to the results of the correlation analysis, the behavior data is divided into positive impact data and negative impact data, and behaviors that have a positive effect on physical health. For example, through analysis, it is found that tracking the target's sleep between 1 pm and 3 pm will make the current blood pressure more stable, thus judging that it has a positive effect on heart rate and blood pressure. When sitting still for a long time, it will cause the diastolic blood pressure of the day to rise, which has a negative impact on blood pressure.

[0047] S4: The positive impact data and the negative impact data have corresponding impact weights, and a suggestion plan is generated based on the impact weights of the positive impact data and the negative impact data and the frequency of occurrence in the behavior data.

[0048] According to the size of the specific impact, the corresponding impact weights of positive impact data and negative impact data can be assigned. The larger the impact weight, the greater the impact of behavioral data on physical data. The size of the impact weight can be determined by linear regression or decision tree regression. For example, the impact weight of a high-sugar diet on blood sugar is -0.8, and the weight of regular exercise on heart rate improvement is +0.6. Then count the frequency of positive and negative behaviors within a month. For example, within a month, a high-sugar diet occurs 5 times, and the number of days with more than 5,000 steps per day is 4 days. Recommendations can be generated according to preset rules. For example, when generating a recommendation, filter out negative impact data with an absolute value of less than 0.5, and then generate a recommendation based on the specific frequency of occurrence, such as taking 80% of the original frequency of occurrence as the recommended number of negative impact data in the recommendation. According to the above preset rules, the description of high-sugar diet in the recommendation is: reduce the frequency of high-sugar diet to no more than 4 times per month.

[0049] S5: Generate a correction strategy based on the suggested solution, reselect a monitoring solution as the second target solution, and continuously obtain the actual behavior data of the tracking target based on the second target solution.

[0050] S6: Analyze actual behavior data to predict the future behavior patterns of the tracked target, and use correction strategies to correct future behavior patterns.

[0051] Corrective strategies are generated based on the recommended solutions through health knowledge rule mapping. For example, according to the above recommended solutions, the generated corrective strategies include high-sugar diets no more than once a week and not sitting for long periods of time between 1pm and 3pm every day. Then, based on the tracking results of the preset time period and the basic data of the tracking target, a more appropriate monitoring plan is selected again as the second target plan. For example, in the second target plan, the camera monitoring frequency is increased to match the number of meals of the tracking target. After that, the actual behavior data of the tracking target is continuously obtained, and predictions are made based on the actual behavior data and historical behavior data. When it is predicted that the tracking target may exhibit negative behaviors in the future, such as prolonged sitting, reminders are issued in advance to avoid the corresponding behaviors of the tracking target.

[0052] By setting up multiple monitoring schemes and obtaining the basic data of the tracking target, the present invention can select the optimal monitoring scheme as the first target scheme according to the personalized health status of the tracking target, thereby realizing health monitoring with stronger pertinence and adaptability. According to the reasonable configuration of monitoring functions and monitoring frequencies, different monitoring needs can be flexibly responded to. On the basis of obtaining physical data and behavioral data, the behavioral data is divided into positive impact data and negative impact data through association analysis, and the corresponding impact weights are assigned to them. Finally, a targeted optimization suggestion scheme can be generated based on the frequency of occurrence of behavioral data.

[0053] The present invention generates a correction strategy based on the recommended solution, and dynamically adjusts the monitoring solution to the second target solution, and continuously optimizes the monitoring method based on real-time data tracking. By analyzing actual behavior data to predict future behavior patterns, and combining correction strategies to intervene in bad behaviors in advance, a closed-loop management from data collection to behavior intervention is achieved, which greatly improves the effectiveness and intelligence level of health management.

[0054] It is particularly noteworthy that the present invention can provide personalized health advice plans by combining the elderly's own physical data and historical behavior data.

[0055] In this embodiment, selecting the second target solution includes the following steps:

[0056] The dynamic statistical features of body data are extracted based on the sliding window method, and the dynamic statistical features are numerically mapped based on the fuzzy algorithm to obtain the fuzzy membership. The fuzzy membership is mapped to rules based on expert knowledge rules to obtain the rule score, and the average score of all rule scores of the same type is calculated. The average score of different types of rule scores is weighted and summed based on the medical history data in the basic data to obtain the comprehensive status score. The comprehensive status score is divided by threshold so as to divide the tracking target into low risk, medium risk and high risk.

[0057] First, missing values ​​are supplemented and duplicate values ​​are removed, and then the sliding window method is used to extract dynamic statistical features. For example, for heart rate, there is a data sequence [78, 80, 83, 87, 90, 92, 88, 85, 82, 79, 76, 74]. A sliding window with a window size of 4 and a sliding step size of 1 is used to extract the average value, and the average data sequence [82, 85, 88, 89.3, 88.8, 86.8, 83.5, 80.5] is obtained. The membership functions are set for low heart rate, normal heart rate and high heart rate respectively. For example, the triangular membership function of normal heart rate is Among them, μ nor (x) is the fuzzy membership of the value x to the normal heart rate, x is the value in the above average data sequence, when x is 85, its fuzzy membership to the normal heart rate is 0.5. The membership functions of low heart rate and high heart rate can be determined based on actual experience and are not limited here.

[0058] After calculation, when x is 85, the fuzzy membership with low heart rate is 0, and the fuzzy membership with high heart rate is 0.25. Based on the knowledge of health experts, the following rules are defined: Rule 1: If the value of low heart rate is high, the risk is low, and the expert score is 1; Rule 2: If the value of normal heart rate is high, the risk is medium, and the expert score is 2; Rule 3: If the value of high heart rate is high, the risk is high, and the expert score is 3. The formula for calculating the rule score is R = μ low (x)·P1+μnor (x)·P2+μ hig (x)·P3, where R is the rule score, μ low (x) and μ hig (x) are the fuzzy membership of x with low heart rate and with high heart rate respectively, P1, P2 and P3 are the scores of rule 1, rule 2 and rule 3 respectively, and the calculation result is 0*1+0.5*2+0.25*3=0.85.

[0059] The rule score of each average value is calculated by the above method, and the average value of all rule scores is obtained. Then, after assigning weights corresponding to different rule scores based on the user's medical history data, the average values ​​of rule scores of different types (heart rate, blood sugar, blood pressure) are weighted and summed to obtain a comprehensive status score. For example, if the user has a history of heart disease, the rule score corresponding to the heart rate is set to 1.5. If there is no history of heart disease, the weight is set to 1. Finally, the risk is divided according to the comprehensive status score, assuming the thresholds are as follows: low risk: comprehensive status score ≤4, medium risk: 4<comprehensive score ≤8, high risk: comprehensive score>8.

[0060] The monitoring plans are divided into low-frequency plans, medium-frequency plans and high-frequency plans. The corresponding monitoring plans are selected as alternative plans based on the risk classification results of the tracking targets. A user feature matrix is ​​constructed based on the physical data and behavioral data of the tracking targets. A solution feature matrix is ​​constructed based on the monitoring frequency, monitoring time period and monitoring type in the alternative plans. The adaptation scores of the alternative plans are calculated based on the user feature matrix and the solution feature matrix. The alternative plan with the highest adaptation score is selected as the second target plan.

[0061] In this embodiment, there are 5 low-frequency solutions, medium-frequency solutions and high-frequency solutions. If the tracking target is judged to be low risk, 5 solutions in the low-frequency solution are selected as alternative solutions. Through this step, on the basis of being able to calculate the physical risk of the tracking target, the number of matching calculations of the monitoring solution is further reduced, reducing the calculation pressure of the system. The following is a simple example to illustrate the above steps. According to the physical data and behavioral data, the following user feature matrix [1500 160092 3] is constructed, where the numerical value represents the maximum heart rate time period of 15:00-16:00, the maximum heart rate is 92, and the heart rate above 90 occurs 3 times between 15:00-16:00. The constructed scheme feature matrix is ​​[1400 1600 6 1600 2400 12], where the values ​​represent 6 monitoring times from 15:00 to 16:00 and 12 monitoring times from 16:00 to 24:00. According to the scheme feature matrix, it is determined that high-frequency monitoring is carried out during the dangerous time period of the tracking target from 15:00 to 16:00, and the corresponding adaptation score is 1. In practice, a more complex matrix needs to be constructed according to the monitoring parameters, and each adaptation score is added to calculate the final result. The specific technical personnel in this field can set it according to experience, which will not be introduced here.

[0062] In this embodiment, the association analysis includes the following steps:

[0063] The missing values ​​of the physical data are supplemented based on the linear interpolation method to obtain the first data, the behavioral data are encoded to obtain the second data, the representative features in the first data are extracted, the interactive features are generated based on the representative features and the second data, the fusion features generated by splicing the interactive features and the second data are analyzed based on linear regression, and the positive impact data and the negative impact data and the corresponding impact weights are obtained.

[0064] For example, for exercise, in the first hour, second hour, third hour and fourth hour, respectively, sitting, light exercise, moderate exercise and heavy exercise are coded as [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1], and then the representative features of each hour are calculated. For example, if the average heart rate in the first hour is 80, the corresponding interactive feature is [80, 0, 0, 0], which is concatenated with the second data to obtain [80, 0, 0, 0, 1, 0, 0, 0], indicating that under the influence of sitting, the average heart rate in this hour is 80. After processing other data in the same way, linear regression analysis is performed on all data to obtain a linear regression equation, and the influence weight is determined based on the linear regression equation. For example, in the linear regression equation Y=Ax1+Bx2 obtained by analyzing the heart rate, the coefficient A of x1 is a positive number, which represents a positive influence, and the coefficient B of x2 is a negative number, which represents a negative influence. The normalized value of each coefficient can be used as the corresponding influence weight.

[0065] In this embodiment, correcting the future behavior pattern includes the following steps:

[0066] Historical data is obtained, which includes behavioral data and environmental data of past dates. Multiple predicted behavioral patterns for future dates are generated based on the historical data and actual behavioral data. The predicted behavioral pattern with the highest probability of occurrence is selected as the future behavior pattern. The correction strategy includes multiple corrective behaviors. When there is a predicted behavior in the future behavior pattern that violates the corrective behavior, a warning reminder is generated to complete the correction of the future behavior pattern.

[0067] Environmental data may have an impact on the behavior pattern of the tracked target. For example, the tracked target may be more inclined to sit at home on rainy days, so additional environmental data is obtained in this embodiment. Historical data can be trained using time series analysis, regression models, or deep learning networks, and multiple possibilities of future behavior patterns can be output by inputting historical behavior data and environmental data. In particular, the model of this embodiment will output multiple predicted behavior patterns and give the probability of occurrence of each predicted behavior model. For example, the output layer uses a softmax neural network model, which will normalize the probability of occurrence of the output result to between 0 and 1. After determining the future behavior pattern, if there is a predicted behavior that violates the corrective behavior, such as the corrective behavior is to reduce the frequency of high-sugar diet to no more than once a week, combined with historical records, it is found that a second high-sugar diet may appear in the future behavior pattern, and a warning reminder is generated in time to correct it.

[0068] In this embodiment, generating a predicted behavior pattern includes the following steps:

[0069] Set date attributes, which include working days, rest days, and holidays. Set date combination rules, establish a prediction model based on historical data, define the dates two days before the future date as the first date and the second date respectively, obtain the date attributes of the first date and the second date and define them as the first attribute and the second attribute respectively, and determine the third date based on the first attribute, the second attribute, and the date combination rules.

[0070] Specifically, the date combination rule is actually a selection rule for the third date. In the date combination rule, when selecting the third date, it is necessary to determine the date attributes of the future date and the previous first and second dates, and select the last historical date that appears in the same order as the first, second and future dates, and select the date that is the same as the second date from the historical dates as the third date. For example, if tomorrow is Wednesday, the first and second dates are Monday and Tuesday respectively, and Monday and Tuesday of this week are both working days, continue to locate to Tuesday of last week. If Wednesday, Tuesday and Monday of last week are all working days, then Tuesday of last week is the third date. If Tuesday of last week is a working day and Monday is a holiday, Wednesday of the week before last is the third date.

[0071] Compare the historical data of the second date and the third date. If the similarity between the two is less than the first threshold, determine the fourth date again based on the date combination rule, continue to compare the historical data with the same date attributes, locate the abnormal data and eliminate it, and then input the remaining historical data into the prediction model to obtain the predicted behavior pattern.

[0072] Based on the above date combination rules, the second date and the third date must have the same date attribute. The historical data of the first date and the third date are obtained, and the similarity of the historical data is compared. If the similarity between the two is less than the first threshold (70%), it indicates that one of the historical data of the first date and the second date does not belong to regular data. Therefore, the fourth date is introduced according to the date combination rule. It is assumed that the Tuesday of the week before the third date is taken as the fourth date, and the similarity between the historical data of the second date, the third date, and the fourth date is calculated. If the result is that the second date is similar to the fourth date, and the third date is not similar to the second date and the fourth date, the historical data of the third date is eliminated as abnormal data, and the historical data of the first date, the second date, and the fourth date are input into the prediction model to obtain the prediction behavior pattern.

[0073] In this embodiment, comparing historical data includes the following steps:

[0074] The two historical data to be compared are converted into a first sequence and a second sequence respectively, and the same behaviors of the first sequence and the second sequence are compared. If the number of the same behaviors is less than a second threshold, the similarity between the first sequence and the second sequence is set to 0. Otherwise, the similarity between the first sequence and the second sequence is calculated based on the overlapping time length of the same behaviors.

[0075] For example, the first sequence and the second sequence are [1, 2, 4, 5, 7, 9] and [1, 4, 3, 5, 9, 6] respectively, and there are behaviors 1, 4, 5, 9 in both of them. Then the number of identical behaviors is 4. If the second threshold is set to 5, the similarity between the first sequence and the second sequence is 0. If the second threshold is set to 3, the time periods of the identical behaviors 1, 4, 5, 9 in the first sequence and the second sequence are continuously obtained to determine the overlapping time periods of the two. The identical behavior 1 in the first sequence and the second sequence appears at 9:00-9:30 and 9:10-9:40 respectively, and the overlapping time length is 20 minutes. Similarly, the overlapping time lengths of other identical behaviors are calculated. Finally, a fixed reference value is set, for example, 480 minutes, and the ratio of the sum of the overlapping time lengths of all identical behaviors to the fixed reference value is used as the similarity. If the sum of the overlapping time lengths of all identical behaviors is 120 minutes, the similarity is 120 / 480*100%=25%.

[0076] The unselected predicted behavior patterns are used as alternative patterns, and a behavior sequence is generated based on the real-time behavior data of future dates. If the mutual exclusivity between the behavior sequence and the future behavior pattern is greater than a second threshold, a new one is selected from the alternative patterns as the future behavior pattern.

[0077] The selected future behavior pattern may not completely match the actual behavior of the user. In fact, a certain behavior of the tracking target may cause corresponding changes in other subsequent behaviors. For example, if the tracking target goes out in a specified time period in the future, its subsequent behavior pattern will change. Therefore, the present invention is also based on the above steps to continuously adjust the future behavior pattern according to the actual behavior of the tracking target. Specifically, on the future date, a behavior sequence is generated by continuously obtaining the actual behavior of the user, and the mutual exclusion between the behavior sequence and the predicted sequence of the future behavior pattern is calculated. The greater the mutual exclusion, the greater the difference between the two. When the mutual exclusion is greater than the second threshold (30%), and in the alternative mode, the one with the smallest mutual exclusion with the current behavior sequence is selected as the future behavior pattern.

[0078] In this step, the mutual exclusion is calculated based on the following method: the default mutual exclusion between the behavior sequence and the prediction sequence of the future prediction model is defined as 0%, and whenever a behavior different from that in the prediction sequence appears in the behavior sequence, its mutual exclusion is increased by 10%.

[0079] In this embodiment, the prediction model includes a local model and a cloud model. The local model extracts comprehensive features of physical data and behavioral data, encrypts the comprehensive features and uploads them to the cloud model. The cloud model generates a predicted behavior pattern based on the comprehensive features.

[0080] Specifically, the comprehensive features include the above-mentioned dynamic statistical features, representative features, etc. By extracting these features locally, the computing pressure on the cloud can be reduced. During the transmission process, encryption can prevent user information leakage. The cloud model is the linear regression model and random forest model introduced above.

[0081] like Figure 2 As shown, the present application provides a big data-based elderly health data analysis system for implementing the above-mentioned big data-based elderly health data analysis method, the system comprising:

[0082] The tracking module is provided with a plurality of monitoring schemes, obtains the basic data of the tracking target, selects a monitoring scheme as the first target scheme based on the basic data, monitors the tracking target based on the first target scheme, and obtains the physical data and behavioral data of the tracking target in a predetermined time period.

[0083] The analysis module performs correlation analysis on the physical data and the behavioral data, and divides the behavioral data that affects the physical data into positive impact data and negative impact data. The positive impact data and the negative impact data have corresponding impact weights. A suggestion plan is generated based on the impact weights of the positive impact data and the negative impact data and the frequency of occurrence in the behavioral data.

[0084] The correction module generates a correction strategy based on the suggested solution, reselects a monitoring solution as the second target solution, and continuously obtains the actual behavior data of the tracking target based on the second target solution.

[0085] The correction module analyzes actual behavior data to predict the future behavior pattern of the tracked target, and corrects the future behavior pattern in combination with the correction strategy.

[0086] The present application provides a computer-readable storage medium having instructions stored thereon, and when the instructions are executed by a processor, the method for analyzing elderly health data based on big data as described above is implemented.

[0087] It should be understood that the various technical features of the above-mentioned embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the various technical features in the above-mentioned embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0088] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for analyzing elderly health data based on big data, characterized in that: Setting multiple monitoring schemes, obtaining basic data of the tracking target, and selecting one of the monitoring schemes as the first target scheme based on the basic data; Monitoring the tracking target based on the first target plan to obtain physical data and behavioral data of the tracking target in a predetermined time period; Performing correlation analysis on the body data and the behavior data, and dividing the behavior data affecting the body data into positive impact data and negative impact data; The positive impact data and the negative impact data have corresponding impact weights, and a suggestion scheme is generated based on the impact weights of the positive impact data and the negative impact data and the frequencies of occurrence in the behavior data; Generate a correction strategy based on the suggested solution, reselect one of the monitoring solutions as a second target solution, and continuously obtain actual behavior data of the tracking target based on the second target solution; The actual behavior data is analyzed to predict the future behavior pattern of the tracking target, and the future behavior pattern is corrected in combination with the correction strategy.

2. The method according to claim 1, characterized in that Selecting the second target solution includes the following steps: Extracting dynamic statistical features of the body data based on a sliding window method, numerically mapping the dynamic statistical features based on a fuzzy algorithm to obtain a fuzzy membership, mapping the fuzzy membership to rules based on expert knowledge rules to obtain a rule score, calculating the average of all rule scores of the same type, and weighted summing the averages of rule scores of different types based on the medical history data in the basic data to obtain a comprehensive status score, and performing threshold division on the comprehensive status score to divide the tracking target into low risk, medium risk and high risk; The monitoring schemes are divided into low-frequency schemes, medium-frequency schemes and high-frequency schemes; the corresponding monitoring schemes are selected as alternative schemes based on the risk classification results of the tracking target; a user feature matrix is ​​constructed based on the physical data and the behavioral data of the tracking target; a scheme feature matrix is ​​constructed based on the monitoring frequency, monitoring time period and monitoring type in the alternative schemes; the adaptation scores of the alternative schemes are calculated based on the user feature matrix and the scheme feature matrix; and the alternative scheme with the highest adaptation score is used as the second target scheme.

3. The method according to claim 1, characterized in that The following steps are involved in the association analysis: The missing values ​​of the physical data are supplemented based on the linear interpolation method to obtain the first data, the behavioral data are encoded to obtain the second data, representative features in the first data are extracted, interactive features are generated based on the representative features and the second data, fusion features are generated by splicing the interactive features and the second data, and the fusion features are analyzed based on linear regression to obtain positive impact data and negative impact data and the corresponding impact weights.

4. The method according to claim 1, characterized in that: Correcting the future behavior pattern includes the following steps: Acquire historical data, wherein the historical data includes the behavioral data and environmental data of past dates, generate multiple predicted behavior patterns for future dates based on the historical data and the actual behavior data, select the predicted behavior pattern with the highest probability of occurrence as the future behavior pattern, the correction strategy includes multiple corrective behaviors, and when there is a predicted behavior in the future behavior pattern that violates the corrective behavior, generate a warning reminder to complete the correction of the future behavior pattern.

5. The method according to claim 4, characterized in that Generating the predicted behavior pattern comprises the following steps: Setting date attributes, the date attributes including working days, rest days and holidays, setting date combination rules, establishing a prediction model based on the historical data, defining the dates two days before the future date as the first date and the second date respectively, obtaining the date attributes of the first date and the second date and defining them as the first attribute and the second attribute respectively, and determining the third date according to the first attribute, the second attribute and the date combination rule; Compare the historical data of the second date and the third date. If the similarity between the two is less than the first threshold, determine the fourth date again based on the date combination rule, continue to compare the historical data with the same date attributes, locate and eliminate the abnormal data, and then input the remaining historical data into the prediction model to obtain the predicted behavior pattern.

6. The method according to claim 5, characterized in that Comparing the historical data comprises the following steps: The two pieces of historical data to be compared are respectively converted into a first sequence and a second sequence, and the same behaviors of the first sequence and the second sequence are compared. If the number of the same behaviors is less than a second threshold, the similarity between the first sequence and the second sequence is set to 0; otherwise, the similarity between the first sequence and the second sequence is calculated based on the overlapping time length of the same behaviors.

7. The method according to claim 4, characterized in that The unselected predicted behavior pattern is used as an alternative pattern, and a behavior sequence is generated based on the real-time behavior data of the future date. If the mutual exclusivity between the behavior sequence and the future behavior pattern is greater than a second threshold, one of the alternative patterns is reselected as the future behavior pattern.

8. The method according to claim 5, characterized in that The prediction model includes a local model and a cloud model. The local model extracts comprehensive features of the body data and the behavior data, encrypts the comprehensive features and uploads them to the cloud model. The cloud model generates the predicted behavior pattern based on the comprehensive features.

9. A big data-based elderly health data analysis system, used to implement the big data-based elderly health data analysis method as described in any one of claims 1 to 8, characterized in that: include: A tracking module, which is provided with a plurality of monitoring schemes, obtains basic data of a tracking target, selects one of the monitoring schemes as a first target scheme based on the basic data, monitors the tracking target based on the first target scheme, and obtains physical data and behavioral data of the tracking target in a predetermined time period; an analysis module, performing correlation analysis on the body data and the behavior data, dividing the behavior data affecting the body data into positive impact data and negative impact data, wherein the positive impact data and the negative impact data have corresponding impact weights, and generating a suggestion plan based on the impact weights of the positive impact data and the negative impact data and the occurrence frequency in the behavior data; A correction module generates a correction strategy based on the suggested solution, reselects one of the monitoring solutions as a second target solution, and continuously obtains actual behavior data of the tracking target based on the second target solution; The correction module analyzes the actual behavior data to predict the future behavior pattern of the tracking target, and corrects the future behavior pattern in combination with the correction strategy.

10. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the method for analyzing elderly health data based on big data as described in any one of claims 1 to 8 is implemented.

Images