Adaptive Impulsive Offset-Free MPC to Handle Parameter Variations for Type 1 Diabetes Treatment

Physiological variations in people with type 1 diabetes constantly change the insulin requirements of patients which, if not compensated, can lead to insulin overdose or insulin insufficiency causing hypoglycemia and hyperglycemia episodes, respectively. Here, an offset-free zone model predictive control (ZMPC) strategy with artificial variables that automatically adjust its penalty parameters is developed by means of new adaptation rules to reduce both types of episodes. The online adaptive tuning is carried out according to the estimation of the plant–model mismatch, the blood glucose value, and its rate of change, producing an aggressive or conservative action depending on the actual situation. In addition, the MPC formulation considers the input of insulin as an impulse instead of a discrete one. The developed method is evaluated in 30 virtual patients of the UVA/Padova simulator and it is compared with an offset-free ZMPC without the adaptation rule. A significant reduction of hypoglycemia episodes is obtained and, for adults, adolescents, and children, a time in normoglycemia range of 87.0%, 67.9%, and 66.1%, respectively, is achieved in a simulation scenario without meal announcement, 30% of parameter variations (simultaneously in several parameters), and sensor noise. The proposed method shows the potential of using information about the estimated mismatch for the MPC tuning rules to compensate the physiological variations. This without requiring complex modifications of the MPC formulation.