INGAR   05399
INSTITUTO DE DESARROLLO Y DISEÑO
Unidad Ejecutora - UE
congresos y reuniones científicas
Título:
A PROBABILISTIC APPROACH FOR CONTROLLING VARIABILITY IN BLOOD GLUCOSE CONCENTRATION
Autor/es:
MARIANO DE PAULA; ERNESTO C. MARTINEZ
Lugar:
Buenos Aires
Reunión:
Congreso; AADECA 2012 – Semana del Control Automático –23º Congreso Argentino de Control Automático; 2012
Institución organizadora:
Asociación Argentina de Control Automático (AADECA)
Resumen:
Insulin-dependent diabetes mellitus is a chronic disease that requires a careful management of insulin infusion rates. In this paper a novel approach for controlling variability in blood glucose is proposed for seeking an optimal control policy in the face of an uncertain dynamics. A near-optimal control policy is developed by combining reinforcement learning with Gaussian processes approximation. Bayesian active learning is used for optimal data selection to guarantee safety and performance. The obtained optimal control policy is compactly represented using Gaussian Processes over a wide range of physiological states. A distinctive advantage of this type of policy approximation is that it can be applied via simple function evaluations facilitates an “on-a-chip” implementation. Results obtained in computational experiments are very encouraging