Invalidation and low-order model set for artificial pancreas robust control design

Abstract

The purpose of this work is to compute a linear parameter-varying (LPV) model set that describes the insulin-glucose dynamics in type 1 diabetes (T1D). This set includes a nominal LPV model and dynamic uncertainty and is amenable to controller design. The nominal model is an LPV control-oriented model previously published by the authors that is (in)validated in this work against the UVA/Padova metabolic simulator. The result is a set of models that is used to design a switched LPV robust controller to account for nonlinearities and variations in insulin sensitivity (SI). Closed-loop responses obtained with the robust controller and a nominal one are compared. Results illustrate the convenience of including robust strategies in designing control laws for an artificial pancreas (AP).