The objective of this thesis is to develop control algorithms for managing blood glucose levels in type 1 diabetic patients, which are based on patient data, in order to obtain personalized algorithms and to create increasingly autonomous devices. Specifically, Model Predictive Control (MPC) is used as the control algorithm, based on the CHoKI (Componentwise Holder Kinky Inference) as a data-driven learning method. The thesis presents different types of CHoKI-based MPC, with different structures, all of which have been tested on virtual patients in the UVA/Padova simulator, accepted by the FDA for preclinical studies. The results are satisfactory, as the various controllers proposed are able to reduce the time spent in hypoglycemia (i.e., when blood glucose levels are below 70 mg/dL), given its short-term danger.
L'obiettivo di questa tesi è quello di sviluppare algoritmi di controllo per la gestione dei livelli di glicemia nel sangue dei pazienti diabetici di tipo 1, basati sui dati del paziente, in modo da ottenere algoritmi personalizzati e creare dispositivi sempre più autonomi. Nello specifico, viene utilizzato il Model Predictive Control (MPC) come algoritmo di controllo, basato sulla CHoKI (Componentwise Holder Kinky Inference) come metodo di apprendimento basato sui dati. Nella tesi sono presentati diversi tipi di CHoKI-based MPC, con strutture diverse e tutti sono stati testati sui pazienti virtuali del simulatore UVA/Padova, accettato dall'FDA per gli studi preclinici. I risultati sono soddisfacenti, in quanto i vari controllori proposti sono in grado di ridurre i tempi in ipoglicemia (ovvero quando i valori di glicemia sono inferiori a 70 mg/dL), vista la sua pericolosità nel breve periodo.
Data-driven Model Predictive Control strategies for blood glucose regulation in Artificial Pancreas
SONZOGNI, Beatrice
2026
Abstract
The objective of this thesis is to develop control algorithms for managing blood glucose levels in type 1 diabetic patients, which are based on patient data, in order to obtain personalized algorithms and to create increasingly autonomous devices. Specifically, Model Predictive Control (MPC) is used as the control algorithm, based on the CHoKI (Componentwise Holder Kinky Inference) as a data-driven learning method. The thesis presents different types of CHoKI-based MPC, with different structures, all of which have been tested on virtual patients in the UVA/Padova simulator, accepted by the FDA for preclinical studies. The results are satisfactory, as the various controllers proposed are able to reduce the time spent in hypoglycemia (i.e., when blood glucose levels are below 70 mg/dL), given its short-term danger.| File | Dimensione | Formato | |
|---|---|---|---|
|
SONZOGNI_BEATRICE_Tesi def.pdf
accesso aperto
Licenza:
Tutti i diritti riservati
Dimensione
7.73 MB
Formato
Adobe PDF
|
7.73 MB | Adobe PDF | Visualizza/Apri |
I documenti in UNITESI sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/20.500.14242/357209
URN:NBN:IT:UNIBG-357209