Research On Short-term Blood Glucose Prediction Model Based On CEEMDAN And ELM

Due to the change of lifestyle,the incidence of diabetic shows younger trend,it has become a disease which affects human health seriously.Inject insulin is good for the glucose control and can reduce the incidence of complications.It is also the basis of artificial pancreas.Blood glucose prediction is not only the key to the control of blood glucose,but also can provide data support for doctors and patients.Therefore,it is of great significance to improve the prediction accuracy and increase the prediction time.Blood glucose prediction is based on the history of blood glucose in human to predict the trend of blood glucose concentration in the future.Based on the research of the blood glucose prediction technology,a short-term prediction model of blood glucose based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and Extreme Learning Machine(ELM)was proposed.In order to improve the accuracy of blood glucose prediction,prediction model uses signal analysis technology and smooths time sequence of blood glucose by means of CEEMDAN.The fluctuation and trend of blood glucose at different scales was decomposed to reduce the non-linearity and non-stationary of the time sequence of blood glucose.A series of blood glucose components with different frequency were obtained.Then,ELMs were built for each IMF and residual component.Finally the all forecasts of ELMs were fused to produce the prediction of blood glucose.On the basis of the short-term blood glucose prediction model of CEEMDAN-ELM,a new method of hypoglycemia warning was designed.The model is verified by 60 cases of type Ⅱ diabetics and the performance of the prediction model is tested by the Clarke Error Grid Analysis and paired t test.The false alarm rate and missing alarm rate were used to evaluate the effect of hypoglycemia.The experiments shows that compared to ELM model and EMD-ELM model,the proposed prediction model of blood glucose based on CEEMDAN-ELM can achieve high accuracy(The 60 cases of diabetics average of RMSE and MAPE is 0.2046 and 2.0855%)for 45 min prediction in advance.The predictive value of blood glucose in patients with 45 min in advance are felled in the A region of the Clarke error grid.The false alarm rate and missing alarm rate of early alarm algorithm of hypoglycemia are 0.77% and 8.71% respectively of 26 diabetics with hypoglycemia events.Short-term prediction of blood glucose based on CEEMDAN and ELM is of great significance to improve the therapeutic effect of diabetes.