A nonlinear blind identification approach to modeling of diabetic patients

Modeling, simulation and control have become effective tools for the treatment of type 1 diabetic patients in the last decades. The availability of reliable models able to predict and/or simulate the behavior of diabetic patients is thus fundamental in this context. Several models, based on first principles or black-box approaches, have been proposed to fulfill this need. However, a common problem to these approaches is that they are not able to recover or to systematically account for the various unmeasured signals which affect a diabetic patient (e.g. food, physical activity, emotions, etc.). In this paper, we propose a blind identification approach, which allows us to derive accurate models of type 1 diabetes patients and to efficiently recover the unmeasured input signals. A simulated example, regarding identification of the blood glucose concentration in type 1 diabetes patients, is presented to demonstrate the effectiveness of the proposed approach.