Nonlinear Model Predictive Control and Artificial Pancreas Technologies

Abstract

A single-hormone artificial pancreas (AP) for people with type 1 diabetes consists of a continuous glucose monitor (CGM), a control algorithm, and an insulin pump for administration of fast acting insulin. In this paper, we describe a control algorithm based on nonlinear model pre- dictive control (NMPC) and demonstrate its performance by simulation using an ensemble of virtual patients. The NMPC is based on: 1) a novel formulation of the objective function separating the computed insulin into basal insulin and bolus insulin; 2) a continuous-discrete time model, where continuous stochastic differential equations describe identifiable insulin- glucose dynamics in the body and the observations by the CGM are at discrete times; 3) a nonlinear filtering and prediction algorithm for the continuous-discrete system that is used both offline for identification of the system and online for state estimation; 4) computationally efficient and robust optimization algorithms for the numerical solution of constrained optimal control problems. The algorithm provides insight into the principles for optimal regulation of the glucose concentration for people with type 1 diabetes.