Physiology-Compliant Empirical Model for Glycemia Prediction

Abstract

This paper is primarily focused on the problem of maximizing glycemia prediction accuracy for the type 1 diabetic subjects, particularly if the linear empirical models are used. The two-input Box-Jenkins model has been preferred since its structure reflects the actual differences in the dynamics of the insulin administration and the carbohydrates intake inputs. Additionally, the basal state of the subject was integrated directly into to model yielding the ability to estimate the basal glycemia within the identification procedure. Since it is common to experience issues with the physiology compliance of empirical models, it was proposed to perform the multi-step-ahead predictive identification of the zero-pole-gain representation with applied constraints. The effect of the excitation signals on the quality of estimated models has also been marginally analyzed, concluding that applying the variable insulin-carbohydrates ratio between individual insulin boluses is a suitable strategy.