Design and testing adaptive predictive control based on global models of endocrine system for diabetic patients

This work presents a new methodology to design the Predictive Functional Control (PFC) algorithm to be implemented in the context of the Artificial Pancreas (AP) for patients with Type 1 Diabetes Mellitus (T1DM). The proposal here is that the parameters of the internal model of PFC be continuously adapted to follow the real patient evolution through an on line identification technique. We call the new PFC as "adaptive PFC" where the internal model parameters are estimated by the recursive estimation based on UD factorization. Additionally, to test the adaptive PFC we use a global data driven model, previously developed, to perform the final tuning (commissioning) to be practically ready to be used in AP implementations. The closed loop responses are tested with the in silico patient obtained through dynamic simulations of the global model (GM) when different meals intake and insulin dosages are given to the patient Nº 5041 from the Center of Diabetes Technology (UVa/USA). The GM is a combination between long and short term models and uses a particular approach based on Kalman Filter to improve the short term predictions. It helps to tracking glycaemia every five minutes to mimic a realistic in silico patient. The comparisons will be done with the regular PFC and the patient without control. The regular PFC has the internal model with constant parameters obtained from the ARX model to isolate the insulin impact from the carbohydrates effects on the blood glucose variations. Finally, the full closed loop simulations, taking into account the constraints related to the insulin pump, are included. The performance of the controllers and the free style of the patient are evaluated by means of the Control Variability Grid Analysis (CVGA). Closing the work, we present the final conclusions based on the comparison results and discuss some future works.