CN112735609A - LSTM model-based old people biological signal health monitoring method - Google Patents

LSTM model-based old people biological signal health monitoring method Download PDF

Info

Publication number
CN112735609A
CN112735609A CN202011638164.9A CN202011638164A CN112735609A CN 112735609 A CN112735609 A CN 112735609A CN 202011638164 A CN202011638164 A CN 202011638164A CN 112735609 A CN112735609 A CN 112735609A
Authority
CN
China
Prior art keywords
model
health
data
lstm
parameters
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202011638164.9A
Other languages
Chinese (zh)
Inventor
朱敬华
邓志豪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang Helian Network Technology Co ltd
Original Assignee
Zhejiang Helian Network Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Zhejiang Helian Network Technology Co ltd filed Critical Zhejiang Helian Network Technology Co ltd
Priority to CN202011638164.9A priority Critical patent/CN112735609A/en
Publication of CN112735609A publication Critical patent/CN112735609A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H80/00—ICT specially adapted for facilitating communication between medical practitioners or patients, e.g. for collaborative diagnosis, therapy or health monitoring
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/70—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for mining of medical data, e.g. analysing previous cases of other patients

Landscapes

  • Engineering & Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Public Health (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Evolutionary Computation (AREA)
  • Biophysics (AREA)
  • Primary Health Care (AREA)
  • Artificial Intelligence (AREA)
  • Pathology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Epidemiology (AREA)
  • Databases & Information Systems (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

The invention discloses an LSTM model-based geriatric biological signal health monitoring method, which adopts an LSTM deep learning model which is related to a time sequence and has a better solution to the problems of gradient elimination and gradient explosion in a long sequence training process. Meanwhile, the invention provides a parameter debugging method in the model based on the monitoring accuracy, and determines the directly applicable parameter value in the model through experimental debugging.

Description

LSTM model-based old people biological signal health monitoring method
Technical Field
The invention belongs to the technical field of artificial intelligence intelligent old people care, and particularly relates to an LSTM model-based old people biological signal health monitoring method.
Technical Field
With the increasing pressure of life, more and more people have sub-health and even chronic disease states, and the health problems of the old people are particularly concerned by people.
At present, most health monitoring tools only carry out local real-time monitoring and display on various health parameters of current users, or transmit data to a mobile phone client through Bluetooth for display, and intelligent analysis and abnormal reminding are not available. Although the artificial intelligence has been applied in the health fields such as biological signals, the methods for performing personalized biological signal analysis and health monitoring of the elderly by using the deep learning technology are few.
The LSTM is a special RNN model as a deep learning model, can learn long-term rules and can effectively solve the time sequence related problem, and the health monitoring data of the invention is related to a time sequence, so that the invention provides an LSTM model-based method for monitoring the health of biological signals of old people, and the method can carry out personalized intelligent health monitoring under the condition of combining time dimension and various health characteristics.
Disclosure of Invention
The invention aims to provide an LSTM model-based geriatric biological signal health monitoring method aiming at the defects of the prior art, collected health biological signal data are used as input, an LSTM deep learning model is used for intelligently analyzing and monitoring the geriatric biological data, and the LSTM deep learning model is an individual monitoring method for different geriatric individuals. According to the invention, through the LSTM model, historical data of the old user are collected, personalized data analysis is carried out, meanwhile, correlation is established on different health biological signal indexes, and finally, intelligent health monitoring is carried out under the condition of combining time dimension and various health characteristics, so that the health abnormity monitoring accuracy of the old is improved. Meanwhile, the invention provides a parameter debugging method in the model based on the monitoring accuracy, and determines the directly applicable parameter value in the model through experimental debugging.
The purpose of the invention is realized by the following technical scheme:
1. an LSTM model-based geriatric biological signal health monitoring method is characterized by comprising the following modules:
(1) a data acquisition module: the method comprises the steps that health biological signals of the old are collected by Internet of things equipment at different time intervals, corresponding multi-dimensional health biological signal original matrix data E are obtained, when the collected signal data meet an enough training set, regularization preprocessing is conducted on an original matrix, and the result is used as input data and is sent to a monitoring module to be operated. In the invention, the definition of ' 0 ' represents unhealthy, 1 ' represents healthy, signals collected in the normal use process of a healthy user are all marked as healthy, and meanwhile, in order to improve the accuracy of training, the method also automatically generates partial unhealthy data by combining with the healthy data and simultaneously takes the partial unhealthy data as the input of a model;
(2) a monitoring module: defining the characteristic dimension of the health signals of the old people collected by the internet of things equipment as n, then m inputs will exist at each time t, and the forward propagation update can be expressed by the following formula:
Figure BDA0002879163550000021
the number of neurons defining the hidden layer is n, the model input is a vector of (1, n + m), and the common weight W represented by equation (1) is a matrix of (n + m) × n. And by analogy of other parameters, the health of the output of the model is a typical classification problem, additional labeling processing is required to be carried out on training set data collected through the Internet of things to serve as the output of the model, and the training is carried out as the input of a loss function.
And performing loss function processing on the final output of the model and the label value, and then realizing model training through the back propagation of the neural network. The loss function adopted by the invention is MSE, and the formula is as follows:
Figure BDA0002879163550000031
according to the gradient descent iterative update principle of back propagation, in combination with the algorithm principle, the LSTM needs to perform back propagation through hidden states h and c, and the following formula is defined:
Figure BDA0002879163550000032
Figure BDA0002879163550000033
define the time of the last step as tau, then
Figure BDA0002879163550000034
Figure BDA0002879163550000035
And then, the weight parameters are deduced forward according to the sequence from t +1 to t, so that all the weight parameters can be obtained.
The model can establish connection from the time dimension and different health data dimensions respectively, biological signals of past time and biological signal data of different types are used as standards for judging whether the model is healthy, and the accuracy of model prediction is greatly improved.
2. A LSTM model-based method for monitoring the health of biological signals of old people is characterized by providing a parameter debugging scheme: the accuracy is an important evaluation index of the evaluation model, the accuracy of the model is calculated by inputting the test set into the trained model and calculating the proportion of the output accurate quantity to the total quantity, and the formula is as follows. The larger the parameter of the hidden layer neuron number in the LSTM model is, the larger the training set is, the more complete the model is, the faster the convergence is, and the higher the prediction rate is. However, too many neurons and training sets cause large computational overhead, and the effect of enhancement is not very obvious. The invention utilizes the test set to carry out dynamic debugging on the two parameters, and determines the optimal parameters in the architecture model of the invention under the condition of ensuring high accuracy and low cost.
Figure BDA0002879163550000036
3. An LSTM model-based health monitoring method for biological signals of old people, which is characterized in that the parameters in the LSTM model suitable for biological signal monitoring of old people in the invention are determined by the debugging method in claim 2 as follows:
(1) the learning rate is 0.0001
(2) The number of iterations is 3000
(3) Hidden layer neuron number of 15
(4) The sample size of the training set is 150
The invention has the beneficial effects that:
1. establishing association for different health biological signals of the old by using an LSTM deep learning model, and then carrying out intelligent health monitoring under the condition of combining a time dimension and multiple biological characteristics, so that the health monitoring accuracy of the old is improved;
2. different from a generalized analysis model, the individual monitoring of different users enables the prediction effect to be closer to the real situation of the users.
3. The method determines the specific parameters of the LSTM model for the health monitoring of the old people, and can be directly applied by utilizing the parameters.
Drawings
FIG. 1 is a schematic diagram of the LSTM model for the health monitoring of biological signals of the elderly according to the present invention;
FIG. 2 is a schematic diagram of model debugging and application of the present invention;
Detailed Description
The following describes embodiments of the present invention in further detail with reference to the accompanying drawings.
As shown in fig. 1 and 2, a method for monitoring health of an aged person biological signal based on an LSTM model is characterized by comprising the following modules:
(1) a data acquisition module: the method comprises the steps that health biological signals of the old are collected by Internet of things equipment at different time intervals, corresponding multi-dimensional health biological signal original matrix data E are obtained, when the collected signal data meet an enough training set, regularization preprocessing is conducted on an original matrix, and the result is used as input data and is sent to a monitoring module to be operated. In the invention, the definition of ' 0 ' represents unhealthy, 1 ' represents healthy, signals collected in the normal use process of a healthy user are all marked as healthy, and meanwhile, in order to improve the accuracy of training, the method also automatically generates partial unhealthy data by combining with the healthy data and simultaneously takes the partial unhealthy data as the input of a model;
(2) a monitoring module: defining the characteristic dimension of the health signals of the old people collected by the internet of things equipment as n, then m inputs will exist at each time t, and the forward propagation update can be expressed by the following formula:
Figure BDA0002879163550000051
the number of neurons defining the hidden layer is n, the model input is a vector of (1, n + m), and the common weight W represented by equation (1) is a matrix of (n + m) × n. And by analogy of other parameters, the health of the output of the model is a typical classification problem, additional labeling processing is required to be carried out on training set data collected through the Internet of things to serve as the output of the model, and the training is carried out as the input of a loss function.
And performing loss function processing on the final output of the model and the label value, and then realizing model training through the back propagation of the neural network. The loss function adopted by the invention is MSE, and the formula is as follows:
Figure BDA0002879163550000052
according to the gradient descent iterative update principle of back propagation, in combination with the algorithm principle, the LSTM needs to perform back propagation through hidden states h and c, and the following formula is defined:
Figure BDA0002879163550000053
Figure BDA0002879163550000054
define the time of the last step as tau, then
Figure BDA0002879163550000055
Figure BDA0002879163550000056
And then, the weight parameters are deduced forward according to the sequence from t +1 to t, so that all the weight parameters can be obtained.
The model can establish connection from the time dimension and different health data dimensions respectively, biological signals of past time and biological signal data of different types are used as standards for judging whether the model is healthy, and the accuracy of model prediction is greatly improved.
A LSTM model-based method for monitoring the health of biological signals of old people is characterized by providing a parameter debugging scheme: the accuracy is an important evaluation index of the evaluation model, the accuracy of the model is calculated by inputting the test set into the trained model and calculating the proportion of the output accurate quantity to the total quantity, and the formula is as follows. The larger the parameter of the hidden layer neuron number in the LSTM model is, the larger the training set is, the more complete the model is, the faster the convergence is, and the higher the prediction rate is. However, too many neurons and training sets cause large computational overhead, and the effect of enhancement is not very obvious. The invention utilizes the test set to carry out dynamic debugging on the two parameters, and determines the optimal parameters in the architecture model of the invention under the condition of ensuring high accuracy and low cost.
Figure BDA0002879163550000061
3. An LSTM model-based health monitoring method for biological signals of old people, which is characterized in that the parameters in the LSTM model suitable for biological signal monitoring of old people in the invention are determined by the debugging method in claim 2 as follows:
(1) the learning rate is 0.0001
(2) The number of iterations is 3000
(3) Hidden layer neuron number of 15
(4) The sample size of the training set is 150
To further illustrate the method of practicing the present invention, an exemplary embodiment is given below. This preferred embodiment is merely illustrative of the principles of the present invention and does not represent any limitation of the present invention.
The biological characteristics of the elderly, namely body temperature, heart rate and walking speed, are assumed to be three, and are respectively represented by an English letter T, R, S in FIG. 2. The above three biometric data of an elderly person are collected over time. After preprocessing the data, it is input into the LSTM model, after each data input, the model is operated according to the method of claim 1 to solve the model parameters, and fig. 2 also shows the state of the model training at time t. After the model training is completed, inputting the data of the test set into the model, and performing model evaluation debugging by using the method of claim 2 to solve the model accuracy.
And respectively changing three parameters of the learning rate, the iteration times and the hidden layer neuron number in the model by the same method until the accuracy is higher than 95% and the calculation cost is minimum, and determining the parameters of the model in the invention.

Claims (3)

1. An LSTM model-based geriatric biological signal health monitoring method is characterized by comprising the following modules:
(1) a data acquisition module: the method comprises the steps that health biological signals of the old are collected by Internet of things equipment at different time intervals, corresponding multi-dimensional health biological signal original matrix data E are obtained, when the collected signal data meet an enough training set, regularization preprocessing is conducted on an original matrix, and the result is used as input data and is sent to a monitoring module to be operated. In the invention, the definition of ' 0 ' represents unhealthy, 1 ' represents healthy, signals collected in the normal use process of a healthy user are all marked as healthy, and meanwhile, in order to improve the accuracy of training, the method also automatically generates partial unhealthy data by combining with the healthy data and simultaneously takes the partial unhealthy data as the input of a model;
(2) a monitoring module: defining the characteristic dimension of the health signals of the old people collected by the internet of things equipment as n, then m inputs will exist at each time t, and the forward propagation update can be expressed by the following formula:
Figure FDA0002879163540000011
the number of neurons defining the hidden layer is n, the model input is a vector of (1, n + m), and the common weight W represented by equation (1) is a matrix of (n + m) × n. And by analogy of other parameters, the health of the output of the model is a typical classification problem, additional labeling processing is required to be carried out on training set data collected through the Internet of things to serve as the output of the model, and the training is carried out as the input of a loss function.
And performing loss function processing on the final output of the model and the label value, and then realizing model training through the back propagation of the neural network. The loss function adopted by the invention is MSE, and the formula is as follows:
Figure FDA0002879163540000012
according to the gradient descent iterative update principle of back propagation, in combination with the algorithm principle, the LSTM needs to perform back propagation through hidden states h and c, and the following formula is defined:
Figure FDA0002879163540000013
Figure FDA0002879163540000014
define the time of the last step as tau, then
Figure FDA0002879163540000021
Figure FDA0002879163540000022
And then, the weight parameters are deduced forward according to the sequence from t +1 to t, so that all the weight parameters can be obtained.
The model can establish connection from the time dimension and different health data dimensions respectively, and the health data of the past time and different health data are used as standards for judging whether the health is good, so that the accuracy of model prediction is greatly improved.
2. A LSTM model-based method for monitoring the health of biological signals of old people is characterized by providing a parameter debugging scheme: the accuracy is an important evaluation index of the evaluation model, the accuracy of the model is calculated by inputting the test set into the trained model and calculating the proportion of the output accurate quantity to the total quantity, and the formula is as follows. The larger the parameter of the hidden layer neuron number in the LSTM model is, the larger the training set is, the more complete the model is, the faster the convergence is, and the higher the prediction rate is. However, too many neurons and training sets cause large computational overhead, and the effect of enhancement is not very obvious. The invention utilizes the test set to carry out dynamic debugging on the two parameters, and determines the optimal parameters in the architecture model of the invention under the condition of ensuring high accuracy and low cost.
Figure FDA0002879163540000023
3. An LSTM model-based health monitoring method for biological signals of old people, which is characterized in that the parameters in the LSTM model suitable for biological signal monitoring of old people in the invention are determined by the debugging method in claim 2 as follows:
(1) the learning rate is 0.0001
(2) The number of iterations is 3000
(3) Hidden layer neuron number of 15
(4) The sample size of the training set is 150.
CN202011638164.9A 2020-12-31 2020-12-31 LSTM model-based old people biological signal health monitoring method Pending CN112735609A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011638164.9A CN112735609A (en) 2020-12-31 2020-12-31 LSTM model-based old people biological signal health monitoring method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011638164.9A CN112735609A (en) 2020-12-31 2020-12-31 LSTM model-based old people biological signal health monitoring method

Publications (1)

Publication Number Publication Date
CN112735609A true CN112735609A (en) 2021-04-30

Family

ID=75608880

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011638164.9A Pending CN112735609A (en) 2020-12-31 2020-12-31 LSTM model-based old people biological signal health monitoring method

Country Status (1)

Country Link
CN (1) CN112735609A (en)

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108597609A (en) * 2018-05-04 2018-09-28 华东师范大学 A kind of doctor based on LSTM networks is foster to combine health monitor method
CN109902558A (en) * 2019-01-15 2019-06-18 安徽理工大学 A kind of human health deep learning prediction technique based on CNN-LSTM

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108597609A (en) * 2018-05-04 2018-09-28 华东师范大学 A kind of doctor based on LSTM networks is foster to combine health monitor method
CN109902558A (en) * 2019-01-15 2019-06-18 安徽理工大学 A kind of human health deep learning prediction technique based on CNN-LSTM

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
LITTLEHAES: "FP,FN,TP,TN与精确率(Precision),召回率(Recall),准确率(Accuracy)", 《HTTPS://BLOG.CSDN.NET/LITTLEHAES/ARTICLE/DETAILS/83278256》 *
刘建平PINARD: "LSTM模型与前向反向传播算法", <HTTPS://WWW.CNBLOGS.COM/PINARD/P/6519110.HTML> *
张德丰: "《TensorFlow深度学习从入门到进阶》", 30 April 2020, 北京：机械工业出版社 *

Similar Documents

Publication Publication Date Title
CN107516110B (en) Medical question-answer semantic clustering method based on integrated convolutional coding
CN110020623B (en) Human body activity recognition system and method based on conditional variation self-encoder
CN113486578B (en) Method for predicting residual life of equipment in industrial process
CN104615983B (en) Activity recognition method based on recurrent neural network and human skeleton motion sequence
CN108304917A (en) A kind of P300 signal detecting methods based on LSTM networks
CN110478883B (en) Body-building action teaching and correcting system and method
CN109993100B (en) Method for realizing facial expression recognition based on deep feature clustering
CN109389171B (en) Medical image classification method based on multi-granularity convolution noise reduction automatic encoder technology
CN110110809B (en) Fuzzy automaton construction method based on machine fault diagnosis
CN111178288B (en) Human body posture recognition method and device based on local error layer-by-layer training
WO2020224433A1 (en) Target object attribute prediction method based on machine learning and related device
Xu et al. Intelligent emotion detection method based on deep learning in medical and health data
CN107203752A (en) A kind of combined depth study and the face identification method of the norm constraint of feature two
CN112668543B (en) Isolated word sign language recognition method based on hand model perception
CN116992779B (en) Simulation method and system of photovoltaic energy storage system based on digital twin model
CN111428481A (en) Entity relation extraction method based on deep learning
CN110210399A (en) Face recognition method based on uncertainty quantization probability convolution neural network
CN112836105A (en) Large-scale student aerobic capacity clustering method based on movement physiological characterization fusion
CN112560948A (en) Eye fundus map classification method and imaging method under data deviation
CN115376638A (en) Physiological characteristic data analysis method based on multi-source health perception data fusion
Niu et al. An indoor pool drowning risk detection method based on improved YOLOv4
Peng Research on Emotion Recognition Based on Deep Learning for Mental Health
Ying Gated recurrent unit based on feature attention mechanism for physical behavior recognition analysis
CN115147768B (en) Fall risk assessment method and system
CN111553311A (en) Micro-expression recognition robot and control method thereof

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20210430

RJ01 Rejection of invention patent application after publication