Closed-loop Control of an Artificial Pancreatic Βeta Cell Using Multi-parametric Model Predictive Control

A model predictive control algorithm was developed for regulating glycemia in subjects with type 1 diabetes mellitus using exogenous insulin infusion to reject disturbances from oral carbohydrate consumption. A clinically validated nonlinear physiological model of glucose metabolism was linearized and served as the basis for the controller. The control law was developed using multi-parametric programming techniques, so that the online optimization problem was reduced to the evaluation of an affine function. The controller was tuned for optimal response to mixed-meals of up to 40 g carbohydrate. The robustness of the controller was then evaluated for mismatch in the size (25%) and timing (15 minutes) of the meal. Due to the computational ease with which this control law was evaluated online, it would be suitable for implementation on a personal digital assistant with limited computational capacity.