Type 1 diabetes is characterized by a lack of insulin production from the pancreas, causing high blood glucose concentrations and requiring external insulin infusion to regulate blood glucose. A novel procedure of "human-friendly" identification testing using multisine inputs is developed to estimate suitable models for use in an artificial pancreas. A human-friendly multisine input signal offers improved identifiability on the dynamics of insulin to glucose, not causing serious deviations from the normal glucose concentration and satisfying insulin delivery pump specifications within acceptable time periods. An integrated formulation of constrained MPC is considered in order to reduce risks of hypoglycemia and hyperglycemia. Furthermore, a set of meal detection and meal size estimation algorithms are developed to improve meal glucose disturbance rejection when incoming meals are unknown. Closed-loop performance is evaluated by simulation studies of a type 1 diabetic individual, illustrating the ability of the MPC-based artificial pancreas strategy to handle measured and unmeasured meals.
展开▼