An intelligent Attentional-GRU-based model for dynamic blood glucose prediction

Continuous glucose monitoring (CGM) and predicting has the problems of strong time-varying, complex nonlinear, and non-stationary. Given this, one kind of intelligent blood glucose prediction model (Attentional GRU method) is proposed that is based on the gated recurrent unit (GRU) with an attentional mechanism. In this hybrid prediction model, the attentional mechanism gives different weights to the input characteristics of GRU, which makes the prediction model more effective in dealing with short-time series glucose monitoring series as input. It is constructed and compared with simple GRU, long-short memory network (LSTM), and support vector regression (SVM). The dynamic blood glucose monitoring and intelligent prediction experiments for live fish (Scophthalmus Maximus) waterless transportation show that the proposed intelligent algorithm has high accuracy and efficiency than the traditional methods. Through the utilization of this advanced method, the key prediction deviation evaluation (RMSE) was averagely and greatly reduced by 26.1%, 29.2%, 28.5% respectively when compares with GRU, LSTM, SVM within the 60 minutes dynamic prediction. It could provide technical references and a smart support tool for the blood glucose dynamic prediction.