Blood Glucose Level Control in Type ۲ Diabetes Using Model-Free Neural Network Predictive Control

Managing blood glucose levels in patients with type ۲ diabetes mellitus remains a significant clinical challenge, particularly due to the nonlinear and patient-specific dynamics of the glucose-insulin system. This paper presents a Neural Network Predictive Controller (NNPC) for administrating insulin delivery by leveraging a data-driven model that predicts future glucose trends without relying on explicit physiological models. The NNPC was trained using simulated glucose-insulin data and evaluated against conventional Sliding Mode Control (SMC) and reinforcement learning-based Normalized Advantage Function (NAF) controllers under realistic conditions, including meal intake scenarios. Results demonstrate that the NNPC effectively maintains glucose levels smoothly within clinically acceptable ranges, while significantly reducing computational complexity and reliance on precise mathematical models compared to existing approaches. These findings suggest that the proposed NNPC framework holds promise for real-time, personalized insulin therapy, potentially enhancing patient safety and treatment outcomes.

کلیدواژه ها:

Type۲ Diabetes Mellitus ، Insulin Delivery ، Artificial Pancreas System ، Model-Free Control ، Neural Network