Research On Analysis And Application Algorithm Of Health Monitoring Time Series Data

A super-aging society is coming,along with the problem of aging,medical care for the elderly needs to be urgently solved.Such patients have a decline in physiological functions with poor self-control capabilities,and are prone to high blood pressure and pressure ulcers,which greatly affects the patients’ physical and mental health in addition with additional medical expenses.Therefore,under the premise of limited medical resources,a comprehensive monitoring system for bedridden diabetics is urgently needed to efficiently monitor the patients’ physiological state in real time.This article is aimed at the health monitoring of bedridden diabetics,and researches on important monitoring issues such as relevant pressure signal data and blood glucose data(including diet and insulin data)acquisition and analysis,patient activity recognition and blood glucose prediction.Based on the actual situation,practical and effective monitoring models are established.The specific content is summarized as follows:1.For the health monitoring related time series data,firstly the intelligent nursing bed and simulator are used to obtain the relevant pressure signal data and blood glucose data with reasonable experiments.Then pre-processing is applied according to the acquisition experiment and the internal characteristics of the data.Finally we set up different scenarios in consideration of the actual situation to get the dataset.And preliminary analysis of the data are adopted to provide support for subsequent algorithms.2.Aiming at the problem of new data and new categories in the case of insufficient data in action recognition based on pressure signal data,an incremental learning model combining sparse auto encoder and random vector functional-linking neural networks is proposed.The data and category incremental model based on the network can be updated quickly with guaranteed accuracy,which fully reflects its practical application value.Aiming at the situation of sufficient data,a domain adaptation network based on conditional adversarial was proposed.The conditional adversarial between the feature extractor and the domain classifier reduces the problems of different data domains between patients with different symptoms and ensures the model’s generalization ability in practical application.3.For blood glucose prediction,the issue of sufficient and insufficient data is mainly considered.In the case of sufficient data,firstly,the convolution with squeeze and excitation is used to fully mine the periodic characteristics and trend changes of blood glucose data,and then combined with the actual blood glucose input,the dynamic feature is further mined through the long short term memory networks with attention.At last,the accurate prediction is achieved.And aiming at the problem that there is only a small amount of data and a gradually increasing data in a new subject in the case of insufficient blood glucose data,a transfer and incremental model based on autoregressive model with exogenous input is proposed.Through the transfer model,the rapid migration and update of the base model on new objects is achieved,and the model is able to become a universal model by preventing negative transfer.Subsequently,iterative incremental updates of the model on its own object are achieved through the incremental model,which can be updated quickly with guaranteed accuracy.