Recent technological advancements in subcutaneous glucose monitoring and sub- cutaneous insulin delivery systems have stimulated the development of closed-loop systems for maintaining normoglycaemia in Type I diabetes. The main obstacle to satisfactory closed- loop control is the presence of disturbances (e.g. meals and physical activity), delays, pump limitation, sensor noise and inter-variability. A Model Predictive Control (MPC) strategy has been applied in a clinical study at the University of Virginia, Padova and Montpellier covering night and breakfast on 20 patients. Night regulation was very satisfactory, while breakfast compensation was slightly worse than with a conventional therapy. This was due to the trade-off in the tuning of the cost function between prompt meal reaction and attenuation of measurement noise. In order to overcome this limitation, in this paper an MPC algorithm added to a conventional therapy is described. In this way all the knowledge incorporated in the (personalized) open-loop therapy is used as a feedforward term while the MPC controller is a feedback correction that guarantees robustness with respect to model uncertainty and incomplete meal knowledge. Finally a procedure for tuning the regulator aggressiveness as a function of few clinical parameters (carbo-ratio and basal insulin delivery) is described.
Model Predictive Control of Type 1 diabetes added to conventional therapy
MAGNI, LALO;TOFFANIN, CHIARA;DE NICOLAO, GIUSEPPE
2011-01-01
Abstract
Recent technological advancements in subcutaneous glucose monitoring and sub- cutaneous insulin delivery systems have stimulated the development of closed-loop systems for maintaining normoglycaemia in Type I diabetes. The main obstacle to satisfactory closed- loop control is the presence of disturbances (e.g. meals and physical activity), delays, pump limitation, sensor noise and inter-variability. A Model Predictive Control (MPC) strategy has been applied in a clinical study at the University of Virginia, Padova and Montpellier covering night and breakfast on 20 patients. Night regulation was very satisfactory, while breakfast compensation was slightly worse than with a conventional therapy. This was due to the trade-off in the tuning of the cost function between prompt meal reaction and attenuation of measurement noise. In order to overcome this limitation, in this paper an MPC algorithm added to a conventional therapy is described. In this way all the knowledge incorporated in the (personalized) open-loop therapy is used as a feedforward term while the MPC controller is a feedback correction that guarantees robustness with respect to model uncertainty and incomplete meal knowledge. Finally a procedure for tuning the regulator aggressiveness as a function of few clinical parameters (carbo-ratio and basal insulin delivery) is described.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
