Blood Glucose Control Based on Hybrid Neural Network and Optimized PID Controller

Type-1 diabetes mellitus (T1DM) is a challenging disorder that requires keeping blood glucose (BG) levels in the euglycemic range for healthy living. To achieve this, we proposed a BG regulation method that integrates a composite neural network glucose prediction model with an optimized proportional-integral-derivative (PID) control algorithm. Firstly, the hybrid neural network model utilizes convolutional neural networks (CNN) and Transformer architecture which enables accurate predictions of future BG trends. Secondly, the PID control algorithm is developed using a particle swarm optimization (PSO) algorithm with standard PID control to set optimum parameters. For validation, in silico experiments have been conducted on ten virtual patients for a 36-hour scenario of the UVA/Padova metabolic simulator. The results revealed an enhancement in BG management. The obtained BG concentration (BGC) values were found in the euglycemic range for 91.5% of the time. Moreover, the BGC values were above 180 mg/dl for 3.8% of the time, the high BG index (HBGI) was found at 1.2 and the blood glucose risk index (BGRI) was also found at 2.5.