Artificial Pancreas Systems: An Integrated Multivariable Adaptive Approach

An artificial pancreas (AP) system with a hypoglycemia early alarm system and adaptive control system based on multivariable recursive time series models is developed. The inputs of the model include glucose concentration (GC) and physiological signals that provide information about the physical activities and stress of the patient. The stability of the recursive time-series models is guaranteed by a constrained optimization method. Generalized predictive control (GPC) is used to regulate GC. Experiments in a clinical setting illustrate the performance of the AP and compare it to open-loop regulation by the patient. Results show that the AP can regulate GC successfully and prevent hypoglycemia in spite of exercise.