Model Research Of Blood Glucose Prediction And Hypoglycemia Early Warning Based On Extreme Learning Machine

Diabetes is a metabolic disease and patients with hyperglycemia and hypoglycemia will cause a series of complications.Repeated hypoglycemia is the main obstacle to reaching blood glucose standards.How to control and treat diabetes effectively has become an important research content.Among the existing medical methods,the use of artificial pancreas to inject insulin into patients is the most effective way to reduce blood glucose.It can achieve accurate blood glucose prediction and hypoglycemia warning,which can not only provide key data for the working system of artificial pancreas,but also provide early warning of hyperglycemia and hypoglycemia,so that patients or doctors can control the unstable events of blood glucose in advance.In this paper,the simultaneous study of blood glucose prediction and hypoglycemia early warning is carried out to realize rapid prediction of hypoglycemia trend and accurate prediction of blood glucose value for patients.In the study of blood glucose prediction,two regression models are established according to the blood glucose data:one is the averaging regression model suitable for any blood glucose data set,the other is the personalized regression model suitable for more blood glucose data sets.In the regression model,elastic net and grey wolf optimization kernel extreme learning machine are used.The root mean square error and mean absolute error are selected as model evaluation indexes.Setting different prediction time for experimental comparison,the curve of personalized model learning is more suitable for the real blood glucose curve of patients.It can reflect the characteristics of patients'own blood glucose,and the prediction is rigorous and effective,so the personalized model in blood glucose prediction is better than the averaging model.In this study,two classification models are established according to the threshold of hypoglycemia early warning:one is the averaging classification model using the critical value of blood glucose as the threshold;the other is the personalized classification model using the characteristics of patients' data to determine the clustering center.In the classification model,Bayesian classification and extreme learning machine are used,the prediction time is 20 minutes and 30 minutes.According to the analysis of experimental results,the personalized model and the averaging model of hypoglycemia early warning have their own advantages and disadvantages,which are complementary to each other.