Short-Term Blood Glucose Prediction Method Based on Signal Decomposition and Bidirectional Networks

For diabetic patients, the monitoring and prediction of blood glucose concentration are crucial aspects of treatment. However, blood glucose concentration is a typical time series data with characteristics such as time variation, nonlinearity, and non-stationarity. In order to mitigate the impact of these characteristics on predictions and enhance the accuracy of blood glucose forecasting, a short-term blood glucose prediction method based on signal decomposition and bidirectional networks (SVBiGL) is proposed.This method employs the Variational Mode Decomposition (VMD) algorithm based on Sparrow Search Algorithm (SSA) to obtain the optimal decomposition of the patient’s blood glucose time series. The decomposed sub-sequences are then predicted using a composite network (BiGL) consisting of Bidirectional Gated Recurrent Unit (BiGRU) and Bidirectional Long Short-Term Memory (BiLSTM) networks. Finally, the predicted sub-sequences are aggregated to obtain the final prediction results.Experimental results demonstrate that the proposed method outperforms BiGRU, BiGL, BiGRU with VMD, and BiGRU with combined SSA and VMD methods in blood glucose prediction at 30-min, 45-min, and 60-min time steps.