Extended adaptive predictive controller with robust filter to enhance blood glucose regulation in type I diabetic subjects

• Three nominal Finite Impulse Response models of the system are used in the control structure.
• Three nominal Finite Impulse Response models of the perturbation effect on blood glucose considered as a meal announcement.
• A robust filter is introduced when the signals are not rich enough for updating the nominal models.
• The glycaemia reference trajectory is shaped by using a rigorous model of a healthy person.
• A variable weighting factor in the cost function is included to prevent dangerous glycaemia excursions out of the allowed limits.

In this paper, an improved adaptive predictive control with robust filter is developed to be applied in an artificial pancreas. Several problems inherent to endocrine systems for diabetic persons have to be tackled such as nonlinearities, long time delays or daily variations of parameters. Three Finite Impulse Response models for insulin input and the same for meal intake (perturbations) corresponding to normal, hyper-hypoglycaemia levels to implement three zones control are taken into account. The glycaemia reference trajectory is shaped from a healthy person response. A variable weighting factor in the cost function is included to prevent dangerous glycaemia excursions out of the allowed limits. Additionally, a noisy blood glucose subcutaneous sensor model is used. This control strategy is tested on 30 virtual subjects from the UVa – Padova Simulator. Simultaneous meals and physiological disturbances are taken into account and the main conclusions are drawn from Control Variability Grid Analysis.