Prediction of Blood Glucose Concentration Based on EMD denoising and LSTM networks

As the number of diabetic patients continues to increase, improving the level of diabetes treatment becomes increasingly important. Blood glucose prediction plays a vital role in diabetes treatment; therefore, it is necessary to further improve the accuracy of blood glucose concentration prediction. Blood glucose data are a kind of sequential data, and the long short-term memory (LSTM) neural network, developed from the recurrent neural network (RNN), is a network structure for dealing with sequential data; therefore, in this study, LSTM networks were used to predict blood glucose concentration. If only LSTM networks are used for prediction, the phenomenon of prediction lag will occur. Therefore, to solve the lag problem and further improve the prediction accuracy, this paper proposes a blood glucose concentration prediction method based on empirical mode decomposition (EMD) denoising and LSTM networks. The final experimental results show that this method can remarkably attenuate the lag phenomenon of prediction and greatly improve the prediction accuracy; therefore, this method can well predict the blood glucose concentration.