Diseño de controladores robustos aplicados a un modelo de páncreas artificial de un paciente virtual con diabetes tipo 1 : diseño y simulación de un controlador basado en redes neuronales aplicado al modelo de glucosa –insulina de hovorka para pacientes con diabetes tipo 1.

Abstract

The study introduced a neural network architecture to design an NARMA-L2 controller that regulated glucose levels in type 1 diabetes patients. The network was trained with various configurations to achieve precise adaptation, addressing each feeding scenario and varying parameters of the average patient in the glucose-insulin Hovorka model. Controller validation proved crucial, using several performance indices such as integral squared error (ISE) and integral absolute error (IAE), in conjunction with the UVA/PADOVA virtual model. Regarding feeding scenarios, the study considered realistic situations involving different meal patterns over three days. Although the NARMA-L2 controller generally outperformed the PID in maintaining lower glucose levels and reducing hyperglycemia duration, it still faced challenges in controlling certain episodes of glycemic variability. Simulation results from both internal simulations and the UVA/PADOVA software highlighted the potential of the NARMA-L2 controller in glycemic management, but also indicated the need for further research to optimize its effectiveness.