Artificial pancreas (AP) systems perform automated insulin delivery to subjects with type 1 diabetes mellitus (T1DM). In this paper, a nonlinear suboptimal controller is designed to make a trade-off between the elimination of hypoglycemia events and the limitation of postprandial hyperglycemia. All the in silico simulations are performed using the distribution version of the UVA/Padova type 1 diabetes (T1D) simulator. The proposed nonlinear AP system is based on an individualized control law which is designed in three steps. At first, a nonlinear model of the glucose–insulin regulatory system is identified based on the data collected from some safe experiments. Then, using the personalized models for all the patients of the simulator and a nonlinear technique called state-dependent Riccati equation (SDRE), suboptimal controllers are designed in which a trade-off between the abilities to correct hyperglycemia and to minimize hypoglycemia is made by considering variable weighting matrices for the controller. Since the SDRE controller has a state-feedback structure, unscented Kalman filter (UKF) is employed to generate estimations for unmeasured state variables from the measured subcutaneous blood glucose level. To assess the performance of the proposed AP system, several scenarios are considered for 33 in silico patients (11 adults, 11 adolescents, and 11 children). The obtained results are analyzed and compared with two other AP systems. Patients’ blood glucose concentrations are maintained in safe levels in all the simulated scenarios and very limited hyperglycemia and no hypoglycemia are observed even in a challenging scenario. The promising results are so encouraging and the proposed AP system is worthy to be tested in vivo.
