Big data-based health data analysis method and system for the elderly, and storage medium

Abstract

The application relates to the technical field of big data processing, and discloses a pension health data analysis method and system based on big data and a storage medium. The method comprises the following steps: monitoring a tracking target based on a first target scheme, acquiring 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 of the behavior data; generating a correction strategy based on the suggestion scheme, reselecting a monitoring scheme as a second target scheme, continuously acquiring actual behavior data of the tracking target based on the second target scheme; analyzing the actual behavior data to predict a future behavior mode of the tracking target, and correcting the future behavior mode in combination with the correction strategy. The application can give a personalized health suggestion scheme in combination with the body data and historical behavior data of the old people.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data processing, and particularly relates to a pension health data analysis method and system based on big data and a storage medium. BACKGROUND

[0002] At present, community pension realizes the collection and management of the living conditions and health information of the elderly through the introduction of intelligent devices, health monitoring systems and community management platforms. However, these technologies are mainly single-function-based, and usually stay at the level of basic health data collection and offline services, and it is difficult to achieve comprehensive analysis and efficient satisfaction of the personalized needs of the elderly.

[0003] As an important part of modern information technology, big data technology has been preliminarily applied in the field of community pension. By integrating and analyzing the health data, living habit 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 a pension personnel health data intelligent analysis method, system, device and medium. The method constructs health data management models of different disease types; obtains the initial health data of the specified collection period of the pension personnel, and obtains the high-risk disease type of the current pension personnel according to the health data management model of different disease types, and then obtains the health management scheme corresponding to the current pension personnel; finally, the behavior tracking data of the current pension personnel is obtained to perform a dangerous behavior state early warning on the current pension personnel. For another example, the Chinese patent document with publication number CN117116480A discloses a pension personnel health data monitoring and analysis system. The system screens out the age interval with the largest proportion of people with the lowest health level of each health indicator, and performs targeted health guidance on the people in the age interval. According to the mean value of different health indicators of each person, a comprehensive health coefficient is calculated, and the result value of the calculation realizes the quantification of the health status of the old people, which is convenient for relevant staff to intuitively understand the comprehensive physical condition of each old person.

[0004] However, the health baseline and behavior habits of different elderly people are different, and the universal model constructed by the above two methods may not adapt to individual differences, and it is difficult to give personalized health recommendation schemes. SUMMARY

[0005] In order to realize comprehensive and accurate service to the health management needs of the elderly, the present application provides a pension health data analysis method and system based on big data and a storage medium.

[0006] In order to achieve the above-mentioned purpose of the application, the present application provides a pension health data analysis method based on big data, comprising:

[0007] A plurality of monitoring schemes are set up to obtain basic data of the tracking target, and one of the monitoring schemes is selected as a first target scheme based on the basic data;

[0008] The tracking target is monitored based on the first target scheme to obtain body data and behavior data of the tracking target in a predetermined time period;

[0009] The body data and the behavior data are analyzed to divide 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 frequency of occurrence in the behavior data;

[0011] A correction strategy is generated based on the suggestion scheme, and one of the monitoring schemes is reselected as a second target scheme, and actual behavior data of the tracking target is continuously obtained based on the second target scheme;

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

[0013] Further, the selection of the second target scheme includes the following steps:

[0014] Dynamic statistical features of the body data are extracted based on a sliding window method, the dynamic statistical features are numerically mapped based on a fuzzy algorithm to obtain fuzzy membership degrees, the fuzzy membership degrees are rule-mapped based on expert knowledge rules to obtain rule scores, an average value of all rule scores of the same type is calculated, and average values of rule scores of different types are weighted and summed based on medical history data in the basic data to obtain a comprehensive state score. The comprehensive state score is threshold-divided 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, a corresponding monitoring scheme is selected as a candidate scheme based on the risk division result of the tracking target, a user feature matrix is constructed based on the body data and the behavior data of the tracking target, a scheme feature matrix is constructed based on monitoring frequencies, monitoring time periods and monitoring types in the candidate schemes, an adaptation score of the candidate scheme is calculated based on the user feature matrix and the scheme feature matrix, and the candidate scheme with the highest adaptation score is selected as the second target scheme.

[0016] Further, the correlation analysis includes the following steps:

[0017] The missing values of the body data are supplemented based on a linear difference method to obtain first data, the behavior data is encoded to obtain second data, representative features in the first data are extracted, interaction features are generated based on the representative features and the second data, the fusion features are generated by splicing the interaction features and the second data, and the fusion features are analyzed based on a linear regression to obtain positive influence data and negative influence data and corresponding influence weights.

[0018] Further, the correction of the future behavior mode includes the following steps:

[0019] The historical data including the behavior data and environment data of past dates are obtained, the future date prediction behavior mode is generated based on the historical data and the actual behavior data, the prediction behavior mode with the highest occurrence probability is selected as the future behavior mode, the correction strategy includes multiple correction behaviors, and when the prediction behavior in the future behavior mode violates the correction behavior, a warning is generated to complete the correction of the future behavior mode.

[0020] Further, the generation of the prediction behavior mode includes the following steps:

[0021] The date attributes including weekdays, weekends and holidays are set, the date combination rules are set, the prediction model is established based on the historical data, the first date and the second date two days before the future date are defined as the first date and the second date respectively, the date attributes of the first date and the second date are obtained and defined as the first attribute and the second attribute respectively, and the third date is determined according to the first attribute, the second attribute and the date combination rules.

[0022] The historical data of the second date and the third date are compared, if the similarity between the two is less than a first threshold, the fourth date is determined again based on the date combination rules, the historical data with the same date attribute are continuously compared, the abnormal data are located and removed, and the remaining historical data are input into the prediction model to obtain the prediction behavior mode.

[0023] Further, the comparison of the historical data includes the following steps:

[0024] The two historical data to be compared are converted into a first sequence and a second sequence respectively, 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 of the first sequence and the second sequence is set to 0, otherwise, the similarity of the first sequence and the second sequence is calculated based on the overlap length of the same behaviors.

[0025] Further, the unselected prediction behavior mode is selected as an alternative mode, a behavior sequence is generated based on real-time behavior data of the future date, and if the mutual exclusivity of the behavior sequence and the future behavior mode is greater than a second threshold value, one of the alternative modes is reselected as the future behavior mode.

[0026] Further, 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, and the comprehensive features are encrypted and uploaded to the cloud model, and the cloud model generates the prediction behavior mode based on the comprehensive features.

[0027] The application provides an elderly health data analysis system based on big data, which is used to realize the above-mentioned elderly health data analysis method based on big data, and the system comprises:

[0028] The tracking module is provided with a plurality of monitoring schemes, acquires 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 acquires body data and behavior data of the tracking target in a predetermined time period;

[0029] The analysis module performs correlation analysis on the body data and the behavior data, divides the behavior data affecting the body data into positive influence data and negative influence data, the positive influence data and the negative influence data have corresponding influence weights, and generates 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;

[0030] The correction module generates a correction strategy based on the suggestion scheme, reselects one of the monitoring schemes as a second target scheme, and continuously acquires actual behavior data of the tracking target based on the second target scheme;

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

[0032] The application provides a computer readable storage medium, and the computer readable storage medium stores instructions, and the instructions are executed by a processor to realize the above-mentioned elderly health data analysis method based on big data.

[0033] The present application can select the optimal monitoring scheme as the first target scheme according to the personalized health status of the tracking target by setting multiple monitoring schemes and obtaining the basic data of the tracking target, and realizes health monitoring with stronger pertinence and adaptability. According to the reasonable configuration of the monitoring function and the monitoring frequency, different monitoring needs can be flexibly responded. On the basis of obtaining the body data and the behavior data, the behavior data is divided into positive influence data and negative influence data through correlation analysis, and the corresponding influence weight is given, and finally the targeted optimization suggestion scheme can be generated based on the frequency of the behavior data.

[0034] The present application generates a correction strategy based on the suggestion scheme, and dynamically adjusts the monitoring scheme to the second target scheme, so as to continuously optimize the monitoring mode on the basis of real-time data tracking. By analyzing the actual behavior data to predict the future behavior mode, and combining the correction strategy to intervene in the bad behavior in advance, the closed-loop management from data acquisition to behavior intervention is realized, and the effectiveness and intelligent level of health management are greatly improved. BRIEF DESCRIPTION OF DRAWINGS

[0035] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating laborious work on the basis of these drawings.

[0036] Figure 1 The schematic diagram of the present application based on big data health data analysis method for the elderly;

[0037] Figure 2 The structure schematic diagram of the present application based on big data health data analysis system for the elderly. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical scheme and advantages of the present application more clear, the following will be further described in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application.

[0039] It can be understood that the terms "first", "second" and the like used in the present application can be used herein to describe various elements, but unless specifically stated, these elements are not limited by these terms. These terms are only used to distinguish the first element from another element. For example, without departing from the scope of the present application, the first xx script can be referred to as the second xx script, and similarly, the second xx script can be referred to as the first xx script.

[0040] As shown in Figure 1 A health data analysis method for the elderly based on big data, comprising:

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

[0042] The monitoring scheme includes monitoring functions and monitoring frequencies of each monitoring function, and the monitoring functions include heart rate, blood pressure, blood oxygen, diet, and drug use. For some functions, different monitoring frequencies are set, such as heart rate, which can be detected every half hour or every ten minutes. Some monitoring schemes do not include the function of monitoring drug use. The tracking target is the old person to be tracked, and the basic data of the tracking target includes disease history, historical blood glucose and blood pressure conditions, and activity preferences. If the tracking target has diabetes, a monitoring scheme including an image recognition function is used to determine the sugar intake of the tracking target by installing a camera in the tracking target's home and obtaining the diet intake category of the tracking target through image recognition. The monitoring scheme selected based on the basic data is used as the first target scheme.

[0043] S2: Monitor the tracking target based on the first target scheme, and obtain body data and behavior data of the tracking target in a predetermined time period.

[0044] S3: Correlation analysis is performed on the body data and behavior data, and the behavior data affecting the body data is divided into positive influence data and negative influence data.

[0045] The predetermined time period is one month, and the corresponding body data and behavior data are collected according to the collection frequency and function of the first target scheme. Heart rate, blood pressure, blood oxygen, body temperature, and sleep quality belong to body data, and step count, exercise duration, exercise content, movement trajectory, diet content, and drug use records belong to behavior data. The above data can be collected by various Internet of Things devices, such as smart watches, smart cameras, and smart medicine boxes.

[0046] The above data is then subjected to correlation analysis, and based on the correlation analysis results, the behavior data is divided into positive influence data and negative influence data, which have a positive effect on physical health. For example, through analysis, it is found that the sleep of the tracking target between 1 pm and 3 pm will make the current blood pressure more stable, thereby determining that it has a positive effect on heart rate and blood pressure. When there is long-term sitting, it will cause the diastolic blood pressure to rise on the same day, which has a negative effect on blood pressure.

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

[0048] According to the size of the specific influence, positive influence data and negative influence data can be given corresponding influence weight, the greater the influence weight, the greater the influence result of behavior data on body data, linear regression or decision tree regression method can be used to determine the size of the influence weight. For example, the influence weight of high-sugar diet on blood sugar is-0.8, and the improvement weight of regular exercise on heart rate is+0.6. Then, the frequency of positive and negative behaviors in a month is counted, such as high-sugar diet appearing 5 times in a month, and the number of days walking more than 5000 steps per day is 4 days. According to the preset rule, the suggestion scheme can be generated, such as filtering out the negative influence data with an absolute value of influence weight less than 0.5 when generating the suggestion scheme, and then generating the suggestion scheme according to the specific frequency, such as taking 80% of the original frequency as the recommended number of negative influence data in the suggestion scheme. According to the above preset rule, the description of high-sugar diet in the suggestion scheme is: reduce the frequency of high-sugar diet to no more than 4 times per month.

[0049] S5: generating a correction strategy based on the suggestion scheme, and reselecting a monitoring scheme as a second target scheme, and continuously acquiring actual behavior data of the tracking target based on the second target scheme.

[0050] S6: analyzing the actual behavior data to predict the future behavior mode of the tracking target, and correcting the future behavior mode in combination with the correction strategy.

[0051] The correction strategy is generated by mapping the health knowledge rule according to the suggestion scheme. For example, according to the above suggestion scheme, the generated correction strategy includes high-sugar diet no more than once a week, and no long sitting between 1pm and 3pm every day. Then, according to the tracking results of the preset time period and the basic data of the tracking target, a more suitable monitoring scheme is selected as a second target scheme, for example, in the second target scheme, the camera monitoring frequency is increased to match the diet frequency of the tracking target. Then, the actual behavior data of the tracking target is continuously acquired, and the actual behavior data and the historical behavior data are predicted, and when it is predicted that the tracking target may appear negative behavior in the future, such as long sitting, a reminder is given in advance, so as to avoid the tracking target from appearing the corresponding behavior.

[0052] The present application can select the optimal monitoring scheme as the first target scheme according to the personalized health status of the tracking target by setting multiple monitoring schemes and acquiring the basic data of the tracking target, and realizes more targeted and adaptive health monitoring. According to the reasonable configuration of monitoring function and monitoring frequency, different monitoring needs can be flexibly coped with. On the basis of acquiring body data and behavior data, the behavior data is divided into positive influence data and negative influence data through correlation analysis, and corresponding influence weight is given, and finally the targeted optimization suggestion scheme can be generated based on the frequency of behavior data.

[0053] The application generates a correction strategy based on a suggestion scheme, dynamically adjusts a monitoring scheme to a second target scheme, and continuously optimizes the monitoring method on the basis of real-time data tracking. By analyzing actual behavior data to predict future behavior patterns and combining the correction strategy to intervene in undesirable behavior in advance, a closed-loop management from data collection to behavior intervention is realized, greatly improving the effectiveness and intelligent level of health management.

[0054] Particularly, the application can give personalized health suggestion schemes in combination with the body data and historical behavior data of the elderly.

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

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

[0057] First, missing values are supplemented and repeated values are removed, and then dynamic statistical features are extracted using the sliding window method. For example, for heart rate, there is a data sequence [78, 80, 83, 87, 90, 92, 88, 85, 82, 79, 76, 74], and the average value is extracted using a sliding window with a window size of 4 and a sliding step of 1 to obtain an average data sequence [82, 85, 88, 89.3, 88.8, 86.8, 83.5, 80.5]. 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 wherein μ nor (x) is the fuzzy membership degree of the value x belonging to normal heart rate, and x is a value in the above average data sequence. When x is 85, the fuzzy membership degree of normal heart rate is 0.5. The membership functions of low heart rate and high heart rate can be determined according to actual experience, which is not limited here.

[0058] After calculation, when x is 85, the fuzzy membership degree of low heart rate is 0, and the fuzzy membership degree of 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, wherein R is a rule score, μ low (x) and μ hig (x) are the fuzzy membership degrees of x to low heart rate and to high heart rate respectively, P1, P2 and P3 are the scores of rule 1, rule 2 and rule 3 respectively, then 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, based on the medical history data of the user, different weights corresponding to different rule scores are assigned, and the average values of rule scores of different types (heart rate, blood glucose, blood pressure) are weighted and summed to obtain a comprehensive state 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, and if there is no history of heart disease, the weight is set to 1. Finally, according to the comprehensive state score, the risk is divided, and the threshold is assumed as follows: low risk: comprehensive state score ≤ 4, medium risk: 4 < comprehensive score ≤ 8, high risk: comprehensive score > 8.

[0060] The monitoring scheme is divided into a low-frequency scheme, a medium-frequency scheme and a high-frequency scheme, a corresponding monitoring scheme is selected as a candidate scheme based on the risk division result of the tracking target, a user feature matrix is constructed based on the physical data and behavior data of the tracking target, a scheme feature matrix is constructed based on the monitoring frequency, monitoring time period and monitoring type in the candidate scheme, an adaptation score of the candidate scheme is calculated based on the user feature matrix and the scheme feature matrix, and the candidate scheme with the highest adaptation score is selected as a second target scheme.

[0061] In this embodiment, there are 5 low-frequency schemes, 5 medium-frequency schemes and 5 high-frequency schemes, if the tracking target is determined to be low risk, 5 schemes in the low-frequency scheme are selected as the candidate schemes. Through this step, on the basis that the body risk of the tracking target can be calculated, the matching calculation times of the monitoring schemes are further reduced, and the system calculation pressure is reduced. The above steps are briefly illustrated as follows. According to the body data and the behavior data, the following user feature matrix [1500 1600 92 3] is constructed, wherein the values represent that the maximum heart rate occurs in the time period of 15:00-16:00, the maximum heart rate is 92, and the heart rate higher than 90 occurs 3 times between 15:00-16:00. The scheme feature matrix is [1400 1600 6 1600 2400 12], wherein the values represent monitoring 6 times at 15:00-16:00 and monitoring 12 times at 16:00-24:00. According to the scheme feature matrix, the dangerous time period of the tracking target is monitored at high frequency at 15:00-16:00, and the corresponding adaptive score is 1. In practice, a more complex matrix needs to be constructed according to the monitoring parameters, and each adaptive score is added to calculate the final result. The specific setting can be made according to the experience of the person skilled in the art, which is not introduced here.

[0062] The correlation analysis in this embodiment includes the following steps:

[0063] The missing values of the body data are supplemented based on the linear difference method to obtain first data, the behavior data is encoded to obtain second data, representative features in the first data are extracted, interaction features are generated based on the representative features and the second data, fusion features are generated by splicing the interaction features and the second data, and the fusion features are analyzed based on linear regression to obtain positive influence data and negative influence data and corresponding influence weights.

[0064] For example, for exercise, sit still, light exercise, moderate exercise and heavy exercise in the first hour, the second hour, the third hour and the fourth hour respectively, and are encoded as [1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0], [0, 0, 0, 1] respectively. Then the representative features of each hour are calculated, for example, the average heart rate of the first hour is 80, and the corresponding interaction feature is [80, 0, 0, 0]. Splicing it with the second data obtains [80, 0, 0, 0, 1, 0, 0, 0], which indicates that the average heart rate in this hour is 80 under the influence of sitting. In the same way, the other data is processed, and linear regression analysis is performed on all the data to obtain a linear regression equation, and the influence weight is determined based on the linear regression equation. For example, the linear regression equation Y = Ax1 + Bx2 obtained by analyzing the heart rate, the coefficient A of x1 is a positive number representing positive influence, and the coefficient B of x2 is a negative number representing negative influence. The normalized value of each coefficient can be used as the corresponding influence weight.

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

[0066] The historical data includes behavior data and environmental data of past dates, a plurality of predicted behavior patterns of the future date are generated based on the historical data and the actual behavior data, the predicted behavior pattern with the highest occurrence probability is selected as the future behavior pattern, the correction strategy includes a plurality of correction behaviors, and a warning reminder is generated to correct the future behavior pattern when there is a predicted behavior that violates the correction behavior in the future behavior pattern.

[0067] The environmental data can have an impact on the behavior pattern of the tracking target, for example, the tracking target may be more inclined to stay at home on a rainy day, so the environmental data is additionally obtained in this embodiment. The historical data can be trained using time series analysis, regression models, or deep learning networks, etc. By inputting the historical behavior data and environmental data, a plurality of possibilities of the future behavior pattern are output. In particular, the model of this embodiment outputs a plurality of predicted behavior patterns and gives the occurrence probability of each predicted behavior model, for example, a neural network model with a softmax output layer normalizes the occurrence probability of the output result to 0-1. After determining the future behavior pattern, if there is a predicted behavior that violates the correction behavior, for example, the correction behavior is to reduce the frequency of high-sugar diet to no more than once a week, and the historical records show that the second high-sugar diet may occur in the future behavior pattern, a warning reminder is generated in time to correct it.

[0068] In this embodiment, the generation of the predicted behavior pattern includes the following steps:

[0069] The date attribute includes weekdays, weekends, and holidays, the date combination rule is set, the prediction model is established based on the historical data, the two days before the future date are defined as the first date and the second date, the date attributes of the first date and the second date are obtained and defined as the first attribute and the second attribute respectively, and the third date is determined according to the first attribute, the second attribute, and the date combination rule.

[0070] Specifically, the date combination rule is actually a selection rule of the third date, in which the date attributes of the future date and the first date and the second date before the first date are determined when the third date is selected, and the last historical date in the same order of occurrence as the first date, the second date and the future date is selected, and the third date is selected from the historical date at the same time as the second date. For example, if tomorrow is Wednesday, the first date and the second date are Monday and Tuesday respectively, and both Monday and Tuesday of the week are working days, continue to locate to last Tuesday, if last Wednesday, Tuesday and Monday are working days, last Tuesday is selected as the third date. If last Tuesday is a working day and Monday is a holiday, the third date is the third date of the week before last.

[0071] By comparing the historical data of the second date and the third date, if the similarity between the two is less than the first threshold value, the fourth date is determined again based on the date combination rule, the historical data with the same date attribute is continuously compared, the abnormal data is located and removed, and the remaining historical data is input into the prediction model to obtain the predicted behavior pattern.

[0072] Based on the above date combination rule, the second date and the third date must have the same date attribute, the historical data of the first date and the third date is obtained, and the similarity of the historical data is compared, if the similarity between the two is less than the first threshold value (70%), it indicates that one of the historical data of the first date and the second date does not belong to the regular data, so the fourth date is introduced according to the date combination rule, assuming that the second date of the third date last week is the fourth date, and the similarity between 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 removed as abnormal data, and the historical data of the first date, the second date and the fourth date is input into the prediction model to obtain the predicted behavior pattern.

[0073] In this embodiment, the comparison of 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, the same behaviors of the first sequence and the second sequence are compared, if the number of the same behaviors is less than the second threshold value, the similarity of the first sequence and the second sequence is set to 0, otherwise, the similarity of 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 respectively [1, 2, 4, 5, 7, 9] and [1, 4, 3, 5, 9, 6], wherein there are behaviors 1, 4, 5, and 9 in both, and the number of the same behaviors is 4, if the second threshold is set to 5, then the similarity of the first sequence and the second sequence is 0. If the second threshold is set to 3, then the time periods of the same behaviors 1, 4, 5, and 9 in the first sequence and the second sequence are obtained, so as to determine the overlapping time periods of the two. The same 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 min. Similarly, the overlapping time lengths of the other same behaviors are calculated. Finally, a fixed reference value is set, for example, 480 min, and the ratio of the sum of the overlapping time lengths of all the same behaviors to the fixed reference value is taken as the similarity. For example, the sum of the overlapping time lengths of all the same behaviors is 120 min, and the similarity is 120 / 480*100% = 25%.

[0076] The unselected predicted behavior mode is taken as an alternative mode, a behavior sequence is generated based on the real-time behavior data of the future date, and if the mutual exclusivity of the behavior sequence and the future behavior mode is greater than the second threshold, then one of the alternative modes is reselected as the future behavior mode.

[0077] The selected future behavior mode does not necessarily completely match the actual behavior of the user. In fact, it is possible that the subsequent other behaviors change correspondingly due to a certain behavior of the tracking target. For example, if the tracking target goes out in a future specified time period, then the subsequent behavior mode will change. Therefore, the application also adjusts the future behavior mode based on the actual behavior of the tracking target according to the above steps. Specifically, on the future date, a behavior sequence is generated by continuously obtaining the actual behavior of the user, and the mutual exclusivity of the behavior sequence and the predicted sequence of the future behavior mode is calculated. The greater the mutual exclusivity, the greater the divergence between the two. When the mutual exclusivity is greater than the second threshold (30%), the one with the smallest mutual exclusivity with the current behavior sequence is selected as the future behavior mode from the alternative modes.

[0078] In this step, the mutual exclusivity is calculated based on the following method. The default mutual exclusivity of the behavior sequence and the predicted sequence of the future prediction model is defined as 0%, and the mutual exclusivity is increased by 10% every time a behavior in the behavior sequence is different from a behavior in the predicted sequence.

[0079] In this embodiment, 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 a predicted behavior mode based on the comprehensive features.

[0080] Specifically, the comprehensive features include the dynamic statistical features, representative features, etc. described above. By extracting these features locally, the computing pressure of the cloud can be reduced. In the transmission process, encryption can avoid leakage of user information, and the cloud model is a linear regression model, a random forest model, etc. described above.

[0081] As shown in Figure 2 The present application provides an elderly health data analysis system based on big data, which is used to realize the above-mentioned elderly health data analysis method based on big data. The system comprises:

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

[0083] The analysis module performs correlation analysis on the body data and behavior data, divides the behavior data affecting the body data into positive impact data and negative impact data, and the positive impact data and negative impact data have corresponding impact weights. The suggestion scheme 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.

[0084] The correction module generates a correction strategy based on the suggestion scheme, and reselects one monitoring scheme as a second target scheme. The actual behavior data of the tracking target is continuously obtained based on the second target scheme.

[0085] 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.

[0086] The present application provides a computer readable storage medium, which stores instructions. When the instructions are executed by a processor, the method for analyzing elderly health data based on big data is realized.

[0087] It should be understood that the technical features of the above-mentioned embodiments can be combined in any way. In order to make the description concise, all possible combinations of the technical features in the above-mentioned embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0088] The above-mentioned are only the preferred embodiments of the present application, and are not used to limit the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for analyzing health data of the elderly based on big data, characterized in that, a plurality of monitoring schemes are set up to obtain basic data of a tracking target, and one of the monitoring schemes is selected as a first target scheme based on the basic data; the tracking target is monitored based on the first target scheme to obtain physical data and behavior data of the tracking target in a predetermined time period; the physical data and the behavior data are analyzed to divide the behavior data affecting the physical data into positive and negative impact data; positive and negative impact data have corresponding impact weights, and a suggestion scheme is generated based on the impact weights of positive and negative impact data and the frequency of occurrence in the behavior data; a correction strategy is generated based on the suggestion scheme, and one of the monitoring schemes is selected as a second target scheme, and actual behavior data of the tracking target is continuously obtained based on the second target scheme; 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; correcting the future behavior pattern includes the following steps: obtain historical data, the historical data including the behavior data and environmental data of past dates, generate a plurality of predicted behavior patterns for future dates based on the historical data and the actual behavior data, select the predicted behavior pattern with the highest occurrence probability as the future behavior pattern, the correction strategy includes a plurality of correction behaviors, when there is a predicted behavior that violates the correction behavior in the future behavior pattern, generate a warning reminder to complete the correction of the future behavior pattern; generating the predicted behavior pattern includes the following steps: set date attributes, including weekdays, weekends and holidays, set date combination rules, establish a prediction model based on the historical data, define the two days before the future date as the first date and the second date, 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 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 a first threshold, then determine the fourth date again based on the date combination rule, continue to compare the historical data with the same date attribute, locate the abnormal data and exclude it, and then input the remaining historical data into the prediction model to obtain the predicted behavior pattern; the date combination rule is actually the selection rule of the third date, in the date combination rule, when selecting the third date, the date attributes of the future date and the first date and the second date before it need to be determined, and the historical date with the same occurrence order as the first date, the second date and the future date is selected, and the historical date with the same time as the second date is selected as the third date. 2.The big data-based health data analysis method for the elderly according to claim 1, wherein, selecting the second target scheme includes the following steps: The dynamic statistical features of the body data are extracted based on a sliding window method, the dynamic statistical features are numerically mapped based on a fuzzy algorithm to obtain fuzzy membership degrees, the fuzzy membership degrees are rule-mapped based on expert knowledge rules to obtain rule scores, the average values of all rule scores of the same type are calculated, and the average values of rule scores of different types are weighted and summed based on the medical history data in the basic data to obtain a comprehensive state score. The comprehensive state score is threshold-divided to divide the tracking target into low-risk, medium-risk, and high-risk; The monitoring scheme is divided into a low-frequency scheme, a medium-frequency scheme, and a high-frequency scheme, a corresponding monitoring scheme is selected as a candidate scheme based on the risk division result of the tracking target, a user feature matrix is constructed based on the body data and the behavior data of the tracking target, a scheme feature matrix is constructed based on the monitoring frequency, the monitoring time period, and the monitoring type in the candidate scheme, an adaptation score of the candidate scheme is calculated based on the user feature matrix and the scheme feature matrix, and the candidate scheme with the highest adaptation score is taken as the second target scheme. 3.The big data-based health data analysis method for the elderly according to claim 1, characterized in that, The correlation analysis includes the following steps: The missing values of the body data are supplemented based on a linear difference method to obtain first data, the behavior data is encoded to obtain second data, representative features in the first data are extracted, interaction features are generated based on the representative features and the second data, fusion features are generated by splicing the interaction features and the second data, and the fusion features are analyzed based on linear regression to obtain positive influence data and negative influence data and corresponding influence weights. 4.The big data-based health data analysis method for the elderly according to claim 1, wherein, The comparison of the historical data includes the following steps: The two historical data to be compared are converted into a first sequence and a second sequence respectively, 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 of the first sequence and the second sequence is set to 0, otherwise, the similarity of the first sequence and the second sequence is calculated based on the overlapping time length of the same behaviors. 5.The big data-based health data analysis method for the elderly according to claim 1, wherein, The unselected prediction behavior mode is taken as a candidate mode, a behavior sequence is generated based on real-time behavior data of the future date, and if the mutual exclusivity of the behavior sequence and the future behavior mode is greater than a second threshold, one of the candidate modes is reselected as the future behavior mode. 6.The big data-based health data analysis method for the elderly according to claim 1, wherein, 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, the comprehensive features are encrypted and uploaded to the cloud model, and the cloud model generates the prediction behavior mode based on the comprehensive features.

7. A big data-based health data analysis system for the elderly, configured to implement the big data-based health data analysis method for the elderly according to any one of claims 1-6. It includes: A tracking module is provided with a plurality of monitoring schemes, acquires 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 acquires body data and behavior data of the tracking target in a predetermined time period. The analysis module performs correlation analysis on the body data and the behavior data, divides 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, generates a suggestion scheme based on the impact weights of the positive impact data and the negative impact data and the frequency of occurrence in the behavior data; The correction module generates a correction strategy based on the suggestion scheme, and reselects one of the monitoring schemes as a second target scheme, continuously acquires actual behavior data of the tracking target based on the second target scheme; The correction module analyzes the actual behavior data to predict a future behavior pattern of the tracking target, and corrects the future behavior pattern in combination with the correction strategy.

8. A computer-readable storage medium having stored thereon instructions, the computer-readable storage medium comprising: The instructions are executed by the processor to implement the method for analyzing health data of the elderly based on big data according to any one of claims 1-6.

Images