Artificial Pancreas: from Control-to-Range to Control-to-Target

Abstract In the last decade, control algorithms designed for Artificial Pancreas (AP) systems were characterized by significant progresses. In particular, the Control-to-Range Model Predictive Control (MPC) showed its effectiveness and safety in several real life studies. Recent studies on model individualization and the enhanced quality of glucose sensors further improved the efficacy of MPC, thus allowing moving from a Control-to-Range to a Control-to-Target approach. In this study, an integral action in the MPC approach (IMPC) is proposed. This ensures beneficial effects in terms of regulation to the target in presence of disturbances such as delays, pump limitation and model uncertainties. The integral action is even more important when model individualization is performed since, during the identification phase, it allows to focus on the identification of the dynamical part of the model rather than to the static gain. The patient models considered in this contribution have been identified through a constrained optimization approach. A procedure for tuning the IMPC aggressiveness by considering both the glucose control performance and the integral of the error with respect to the target is described. Finally, in silico experiments are presented to assess the effectiveness of the proposed IMPC.