Blood Glucose Regulation Modelling and Intelligent Control

A Master of Science thesis in Mechatronics Engineering by Nabeel Nasar entitled, “Blood Glucose Regulation Modelling and Intelligent Control”, submitted in December 2024. Thesis advisor is Dr. Lotfi Romdhane and thesis co-advisor is Dr. Mohammad Jaradat. Soft copy is available (Thesis, Completion Certificate, Approval Signatures, and AUS Archives Consent Form).

Abstract

In this thesis, the Reinforcement Learning - Proportional Integral Derivative (RL-PID) and the Particle Swarm Optimization based Reinforcement Learning – Adaptive Proportional Integral Derivative (PSO-RL-PID) Controllers are used to regulate the blood glucose regulation process. The proposed reinforcement learning algorithm used to train the system is called Twin Delayed Deep Deterministic (TD3) Policy Gradient method. The advantages include delayed policy updates, dual critic function evaluation and consideration of noise, which accounts for measurement noise from the blood glucose sensor. The Bergman Minimal (BM) Model, which is considered a Low Complexity (LC) model and the Li & Kuang (LK) Two Delay Diabetic Patient Model, which is of Medium Complexity (MC) model, are used for the regulation process. The Reinforcement Learning Agent is used to tune the PID parameters. After training, the controller’s performance is analyzed by adding external disturbances simulating meal intakes and a constant band-limited white noise. The results indicate that the proposed controllers are able to regulate the blood glucose level while ensuring that the blood glucose level remains within safe limits. Additionally, performance comparison between the proposed controller and other control techniques is analyzed. The results indicate increased performance and lower error between the desired blood glucose reference and the measured blood glucose reference values in comparison to other AI- based controllers.

DSpace URI

https://hdl.handle.net/11073/25807

Collections

Masters Theses

Show all metadata