Model predictive control for infusion pump insulin delivery

The authors present an algorithm for controlling blood glucose in the Type I diabetic patient using model predictive control (MPG) of a closed-loop insulin infusion pump. The controller is designed from a detailed 19th order model of the human glucose-insulin system. Input constraints, based on insulin pump and physiological requirements, are enforced. The controller is challenged with a hyperglycemic diabetic patient (setpoint tracking), and a controlled diabetic ingesting a meal (unmeasured disturbance rejection). To improve both robustness and performance, a new controller using MPC with state estimation is synthesized. Compared to the non-state estimating controller, the new controller results in a 40% improvement in overshoot, to a maximum hypoglycemic deviation of 7.9 mg/dl, and a 23% decrease in settling time. These results demonstrate the potential use for this algorithm in a closed-loop insulin infusion pump