Auto-Tuning for Model Predictive Controllers in Patients with Type 1 Diabetes

Current proposals in glucose control using closed-loop systems have focused on maintaining blood glucose in a safe range, this against disturbances as meal intake or exercise. Due of its clinical performance, Model predictive control has been positioned as one of the most used control algorithms in the artificial pancreas. However, the personalization of this control strategy is an unresolved issue and little addressed in the literature. In this work, an auto-tuning methodology for MPC controller on type 1 diabetes patients is presented. The tuning is done through the nelder-mead method to find the controller's parameters that maximizes the time inside the normoglycemia range (70 - 180 mg/dl). It is chosen as variables to tune the weighting coefficient of the output and input, the predictive horizon, and the minimum and maximum values of the target zone. The results obtained show that auto-tuning methodology allow an increase up to 54.43% of time in normoglycemia, with an average increase up to 17.21%.