The Research Of Grey Prediction Model In The Prediction Of Blood Glucose

With the continuous improvement of people’s living standards, the growth number of diabetic patients caused great harm to human health. And the main purpose of the clinical treatment of diabetes is to keep stable blood sugar concentration. If the future blood glucose concentration can be predicted, doctors and patients can take effective measures to stabilize the blood glucose level before the accident of hyperglycemia and hypoglycemia happens. This will largely reduce the harm to the patients with unstable blood glucose concentration. In a word, the research which is used to predicting the blood sugar concentration of diabetes is particularly important.At present, there are two directions which are used to predict the blood sugar concentration of diabetes. One is based on physiological model of human body, we need to consider the external factors that impact the change of blood glucose concentration with diabetes, such as diet, drugs injection, mood, movement, etc. In order to achieve the stable blood sugar concentration, it needs to control the blood sugar concentration with diabetes. But the prediction algorithm is complex and the prediction results have a delay. The other is only to use the patient’s history blood glucose values, to establish proper mathematical models to predict the future blood sugar levels, and to achieve a simple and efficient prediction effect.In this thesis, firstly, we researched a variety of blood sugar prediction model that based on the historical blood glucose data with diabetes. For example, the artificial neural network prediction model, the extreme learning machine prediction model, the support vector machine prediction model, the AR prediction model, the fusion of ARIMA and RBFNN prediction model, etc. And the grey prediction model is applied to the field of blood sugar prediction for the first time. Secondly, the blood glucose prediction GM(1, 1) model is proposed based on the grey theory. Two aspects are improved on the basis of the traditional GM(1, 1) model. On the one hand, in order to weaken the randomness of original time series, we preprocessed the acquired blood sugar data so that the data curve becomes smoother. On the other hand, the metabolic algorithm is used. Continually adding new data, deleting the old data, to establish the current blood glucose training set and get the best prediction value of blood sugar. Finally, blood glucose data of 50 cases will be used to verify the prediction model based on GM(1, 1) and be compared with the AR model. The results show that blood sugar prediction GM(1, 1) model has a better effect than the AR prediction model on predicting the future blood sugar levels. Especially it has good prediction accuracy about the postprandial blood glucose based on the blood sugar prediction GM(1, 1) model.