Neural Network-Based Optimal Control for Glucose Regulation in a Simplified Diabetic Model

Abstract

This paper investigates the application of neural networks for approximating optimal control strategies in regulating blood glucose levels in a simplified model of glucose-insulin dynamics. A linear model of glucose-insulin interaction is used, and an optimal control problem is formulated to minimize deviations from a target glucose level while penalizing excessive insulin infusion. Training data for the neural network is generated by numerically solving the optimal control problem. A feedforward neural network is trained on this data to approximate the optimal control policy. The performance of the neural network controller is evaluated through simulation and compared against the directly calculated optimal control, demonstrating the potential of neural networks for personalized glucose regulation. The limitations of this simplified approach and directions for future research are also discussed.