Data-driven polynomial MPC and application to blood glucose regulation in a diabetic patient

The majority of control design approaches assume that an accurate first-principle model of the system to control is available. However, in many real-world applications, deriving an accurate model is extremely difficult, since the system dynamics may be not well known and/or too complex. In this paper, a polynomial model predictive control (PMPC) approach for nonlinear systems is presented, relying on the identification from data of a polynomial prediction model. The main advantages of this approach over the standard methods are that it does not require a detailed knowledge of the plant to control and it is computationally efficient. A real-data application is presented, concerned with regulation of blood glucose concentration in a type 1 diabetic patient. This application shows that the PMPC approach can be effective in the biomedical field, where accurate first-principle model can seldom be found.