Method for predicting physique change trend of natural person by using multi-dimensional wearable sensor data
Through the initial analysis of correlation analysis, combined with GM (1,N) and LSTM models, a prediction model of intelligent wearable data is constructed, which solves the problem of low data processing efficiency of intelligent wearable devices on cloud servers, and improves prediction quality and efficiency.
Patent Information
- Application Number
- CN202410986538.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-19
AI Technical Summary
When existing smart wearable devices process health data on cloud servers, they face problems such as large data parameters, large data volume and poor quality, resulting in low prediction quality and low utilization rate.
The initial table of correlation analysis combined with GM (1,N) and LSTM models is used to compare and fusion each other through the diverse time-system prediction systems. The Keras library is used to build a single data series and overall prediction model of correlation parameters, which reduces the operation dimension and improves the operation efficiency. .
It improves the prediction quality of smart wearable data, improves data processing efficiency, solves the problems of large amounts and poor quality, and achieves a more accurate prediction of changes in natural human physique trends.
Smart Images

Figure FT_1 
Figure FT_2 
Figure SMS_1
Abstract
Description
Technical field:
[0001] The present invention discloses a service cloud, which performs correlation analysis and prediction on the data set according to the initial correlation analysis table (when more than two data are correlated, according to the maximum correlation dimension) for the smart wearable data that has been interpolated and standardized, and comprehensively predicts the change trend of the physique of the data provider (natural person). The cloud server uses the current real-time health data of the smart wearable (blood sugar, blood pressure, heart rate, blood oxygen, exercise data, sleep, skin humidity, temperature, video, picture, location information, etc.), related static (statistical data, age, gender, body shape, health history data released in the region), tongue image, face image, sound, etc. to integrate and do a good job in the early data management work.
[0002] A front-end subsystem in the health service management system of the present invention belongs to the field of cloud technology for health information management supported by cloud computing and other technologies, and can solve the needs of existing smart terminal data to quickly process data and improve prediction quality when the cloud server processes these data on a large scale. Through multiple system predictions, mutual proofreading, and fusion of unified methods, the current smart wearable data parameter dimensions are solved, the data volume is large, and the quality is poor compared with professional medical equipment, and the utilization rate is low. Background technology:
[0003] Data source: Currently, smart wearables are basically equipped with health data sensors, and can use the functions of their own main chips to provide relevant data transmission interface protocols to provide relevant health data to the manufacturer's server or third-party public cloud.
[0004] Number of data: as few as three and as many as ten.
[0005] Prediction data quality: It is necessary to adopt multiple correlation models, data correlation fusion, mutual comparison and improvement.
[0006] Related technologies: LSTM model, GM (1,1) model, relational database, Keras open source library, all of which are mature technologies. Comparison algorithm to compare whether the value is within a reasonable range, LSTM model, The GM(1,1) model already has relevant formulas and reference programs. Cloud computing and cloud database technologies have been basically perfected. After inquiry, there is currently no system that can analyze the correlation of multiple physiological indicators of natural persons, compare and integrate multiple time series prediction systems. In particular, there is no algorithm that uses correlation analysis tables to reduce the calculation dimension, improve calculation efficiency, and save transportation capacity.
[0007] References: The correlation between data in each dimension can be found in some relevant materials. At the same time, there is no report that is unified by experts and certified from multiple dimensions. Under the guidance of experts, we used part of the data and deep algorithms, and on the basis of verification, we formed a correlation analysis table for conventional data of smart wear and a correlation analysis table after the fusion of conventional data of smart wear and normal data. The correlation relationship of some dimensional data can be seen in various treatment guidelines and needs to be comprehensively sorted out.
[0008] The key is how to combine existing technologies to improve the quality of data correlation analysis and predictive data.
[0009] Patent Search: CN110322947A 2019.06.14 "A method for recommending exercise prescriptions for elderly people with hypertension based on deep learning". This paper discusses a single-dimensional prediction method.
[0010] Dissertation: "Development of Elderly Fall Detection System Based on Multimodal Information Fusion", Pi Liming. University of Chinese Academy of Sciences (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences). Master's electronic journal publication information: Year and Issue: 2019 Issue 02. This paper proposes the changes in heart rate, blood pressure, body temperature, respiratory rate, and blood oxygen of natural people after a fall. "Research on Elderly Health Monitoring System Based on Multi-sensor Information Fusion", Zhu Tianyang. Beijing University of Chemical Technology. Doctoral Electronic Journal Publication Information: Year and Issue: 2022 Issue 03. The main research work of this paper includes: (1) The relationship between the four vital signs of body temperature, blood pressure, heart rate, and blood oxygen and the gender, Correlation study of factors such as age and measurement time.
[0011] In summary: Currently, no search results are found: With the help of the correlation analysis table, use the grey theory GM (1, N), LSTM model for double prediction, and then use it to generate reports or patents. Summary of the invention:
[0012] According to the suggestions of relevant experts and simulation analysis, the preliminary smart wear data correlation analysis tables 1-1 and 1-2 are formed.
[0013] According to the suggestions of relevant experts and simulation analysis, after adding static data to the smart wear data, the preliminary table 2 of the smart wear data association analysis was formed.
[0014] Based on the smart wear data association analysis of Table 1-1 and Table 1-2, a data set is constructed, and a single data series prediction model of these associated parameters is formed by using GM (1.N), LSTM model, and Keras according to the maximum number of associated parameters.
[0015] Based on the smart wear data association analysis, Table 2, after adding static data, the data set is constructed, and the GM (1.N), LSTM model, and Keras are used to form the overall prediction model of the associated data series of these associated parameters according to the maximum number of associated parameters.
[0016] According to Table 3-1, Table 3-2, Table 3-3, Table 3-4, Table 3-5 and Table 3-6, for values that exceed the reference range of Table 3-1, Table 3-2, Table 3-3, Table 3-4, Table 3-5 and Table 3-6, set the prompt bit to "different" to prepare for the subsequent data display. Specific implementation method:
[0017] Step 1: Rent a database on Alibaba Cloud.
[0018] Step 2: Create relevant forms for natural person smart wearable data.
[0019] Step 3: Establish relevant security mechanisms.
[0020] Step 4: Open the server address and corresponding port on the cloud.
[0021] Step 5: Deploy the application system on the cloud server. Currently, there are four standard templates: Android, Apple, Huawei, and specific wristbands.
[0022] Step 6: Establish the relationship between user name ID, device ID, bound user ID and device ID.
[0023] Step 7: Connect with the device according to the manufacturer's white paper. Sign an authorization and service contract with the user (the same contract for the entire platform).
[0024] Step 8: Under the unified coordination of the upper module, it is integrated with other modules to complete the comprehensive application of smart wear data. This module can be independently packaged and run independently based on certain data. Description of the drawings:
[0025] Figure 1 : Functional module diagram provided in the implementation of the present invention.
[0026] Figure 2 : Data flow diagram provided in the implementation of the present invention.
[0027] Table 1-1, Table 1-2: Preliminary correlation analysis of smart wearable data (no static data dimension) Table 1-1, Table 1-2.
[0028] Table 2: Preliminary correlation analysis of smart wearable data (adding static data dimension) Table 2.
[0029] Table 3-1, Table 3-2, Table 3-3, Table 3-4, Table 3-5, Table 3-6: Blood sugar, blood pressure, Heart rate, blood oxygen, exercise data, sleep, skin humidity, skin temperature and body shape reference values.
[0030] Related formulas. Table 1-1 Parameter association table Table 1-2 Parameter association table Table 2 Preliminary table of correlation analysis of smart wear data (adding static data dimension) Table 3-1: Reference values for blood sugar, blood pressure, heart rate, blood oxygen, exercise data, sleep, skin humidity, skin temperature and body type Table 3-2 Table 3-3 Table 3-4 Table 3-5 Table 3-6 Related formula: (I) GM(1,1) formula 1. Gray Generation 2. Constructing the GM(1,1) model 3. Solving model parameters 4. Assume the original data sequence is: x (0) ={x (0) (1),x (0) (2),...,x (0) (n)} (1) Accumulated generation: (2) Modeling: (3) Parameter estimation (least squares method) Where B and Y are defined as follows: (4) Model solution: (5) Restore the prediction results: (II) Construction process and mathematical formula of GM(1,N) model 1. Data Collection: Collect dependent variables: x(t) and N independent variables u i (t), where t = 1, 2, ..., n is the time point of the time series. 2.- Accumulation (1-AGO): Accumulate all data sequences once to generate a new sequence: Here x (1) and Represent the cumulative sequence of the dependent variable and the i-th independent variable respectively. 3. Constructing grey model: According to the grey system theory, the differential equation for constructing the GM(1,N) model is as follows: Among them, a and b i are the model parameters to be estimated, and b0 is a constant term. 4. Build data matrix and least squares estimation: Construct data matrix B and vector A: Use the least squares method to solve for the parameters a,b0,b1,...,b N : 5. Model solving and restoration: According to the obtained parameters, the prediction model is restored using the difference equation. For each time point t, the predicted value It can be obtained by solving differential equations. 6. Error check: Necessary error tests are performed on the model to evaluate its prediction accuracy and applicability.
Claims
1. Based on the suggestions of relevant experts and simulation analysis, the preliminary tables 1-1 and 1-2 of smart wearable data association analysis were formed.
2. According to the suggestions of relevant experts and simulation analysis, after adding static data to the smart wear data, the preliminary table 2 of the smart wear data association analysis was formed.
3. Relying on the smart wear data association analysis of Table 1-1 and Table 1-2, construct a data set, and use GM (1.N), LSTM model, Keras, according to the maximum number of associated parameters. A method of forming a single data series prediction model of these associated parameters.
4. Relying on the smart wear data association analysis preliminary table 2, on the basis of smart wear data, after adding static data, construct a data set, use GM (1.N), LSTM model, Keras, according to the maximum number of associated parameters. Form a method of the overall prediction model of the associated data series of these associated parameters.
5. According to Table 3-1, Table 3-2, Table 3-3, Table 3-4, Table 3-5 and Table 3-6, for values that exceed the reference range of Table 3-1, Table 3-2, Table 3-3, Table 3-4, Table 3-5 and Table 3-6, set the prompt bit to "different" to prepare for the subsequent data display.
