Deep learning fuzzy immersion and invariance control for type-I diabetes.

Area of Science:

• Biomedical Engineering
• Control Systems
• Artificial Intelligence

Background:

• Type-I diabetes management requires precise glucose regulation.
• Existing glucose-insulin metabolism models often overlook inherent uncertainties.
• External factors like meals and illnesses significantly impact glucose dynamics.

Purpose of the Study:

• To propose a novel glucose regulation strategy for type-I diabetes patients.
• To address uncertainties in glucose-insulin metabolism.
• To develop a robust system capable of handling dynamic perturbations and meal effects.

Main Methods:

• Utilized the Immersion and Invariance (I&I) theorem to derive adaptation rules for unknown parameters.
• Developed a deep learned type-II fuzzy logic system (T2FLS) for error compensation and stability.
• Tuned the T2FLS using singular value decomposition (SVD) and adaptive rules derived from stability analysis.
• Validated the approach using the modified Bergman model (BM), incorporating meal effects and random perturbations.

Main Results:

• The proposed method demonstrated effective glucose regulation, achieving desired levels within a short time.
• Simulations confirmed the approach's superiority compared to other methods.
• The system maintained glucose levels within the target range for over 99% of the time across various diabetic conditions.

Conclusions:

• The novel approach offers a robust and effective solution for glucose regulation in type-I diabetes.
• The integration of I&I theorem and T2FLS successfully handles metabolic uncertainties and external disturbances.
• The method shows significant promise for improving patient outcomes in diabetes management.