Research On Glucose Prediction Model And Hypoglycemia Alarm Technology

Diabetes and its complications have become one of the major diseases affectinghuman health. In recent decades, the number of diabetic patients is further increased,and the incidence tends to be younger. Long-term diabetes will bring series ofcomplications, even life-threatening. Therefore, how to accurately predict bloodglucose, control blood glucose fluctuations in time, will be the main task of thetreatment of diabetes. Although the control of blood glucose can prevent or delaycomplications, the risk of hypoglycemia has increased. Hypoglycemia can cause braincell damaged, coma and even death, so timely prediction of hypoglycemia, especiallyasymptomatic hypoglycemia, is very important in the treatment of diabetes.In the area of research on glucose prediction and hypoglycemia alarm, twoprediction models has proposed. A prediction method is based on physiology model.Another one is based on the data model. Because the physiological mechanism ismore complex, and many factors affect blood glucose, it is difficult to establishaccurate predictive model. The prediction model based on data-driven is completelydependent on the history of blood glucose data. But the accuracy of prediction anduniversality need to be improved.In this paper, a new glucose prediction algorithm, adaptive prediction algorithm,which is an improved algorithm based on Kalman filter and autoregression model, isproposed. First, we use the Kalman filters to reduce the noise which interferences theglucose signals; then we analyze the non-stationary of the glucose signals and use theadaptive forgetting factor(switch variable) AR model to predict the future blood sugar,in the model, the historical glucose data, meals, sports and some other events areconsidered. At last, in order to determine the superiority of prediction performance,we choose RMSE and SSGPE as the predictor effects, and the proposed method hasbeen compared with the traditional AR model on50subjects’ CGMS data.Finally, based on adaptive prediction model, it researches the hypoglycemia warning technology. According two indicators of Sensitivity (Sens), the false positiverate (FPR) to assess hypoglycemia warning system, early warning technology couldtimely discover hypoglycemia, provide evidence of taking the next step for doctors.