Long-term prediction for T1DM model during state-feedback control

Avoiding low glucose concentration is critically important in type-1 diabetes treatment. Predicting the future plasma glucose levels could ensure the safety of the patient. However, such estimation is no trivial task. The current paper proposes a predictor framework which stems from Unscented Kalman filter and works during closed-loop control, that can predict hazardous glucose levels in advance. Once the blood glucose concentration starts to rise, the predictor activates and estimates future glucose levels up to 3 hours, confirming whether the controller can endanger the patient. The capabilities of the framework is presented through simulations based on the SimEdu validated in-silico simulator.