Robust nonlinear model predictive control of diabetes mellitus

Abstract

The purpose of this paper is to present a Robust Nonlinear Model Predictive Control controller design opportunity and the results of three in silico test scenarios, where a nonlinear glucose model had to be controlled, and a desired blood glucose level had to be maintained. The chosen glucose model was a two compartmental, nonlinear model with time delay whose parameters were burdened with uncertainty. During the three test scenarios the controller performed well. It could keep the blood glucose level in the desired range without dangerous undershoots. In the third test scenario, during the simulation of 28 full days, 80% of the daily extremes lied between 5,5–10 mmol/l. The performance and computational bounds that are present at the moment are addressed and possible solutions are given at the end of the paper.