Insulin infusion rate control using information theoretic–based nonlinear model predictive control for type 1 diabetes patients

Abstract

It has been proven that model predictive control (MPC) is an efficient method for closed-loop insulin delivery in clinical studies. This paper aims to design an observer-based fractional-order nonlinear MPC for type 1 diabetes mellitus (T1DM) patients. It is assumed that the proposed model is nonlinear and contains parametric uncertainty. To estimate unknown states, optimal non-fragile H∞ observer is designed for Lipschitz nonlinear fractional-order systems including parametric uncertainty and the existence of input disturbance. The min–max optimization-based robust fractional model predictive control (RFMPC) has been presented in the following for insulin delivery. Since sensor noise of continuous monitoring of interstitial glucose concentration is considered non-Gaussian, the performance of the proposed controller is improved under non-Gaussian measurement noise by selecting a proper cost function based on generalized correntropy, and as a contrast, the performance of the mean square error (MSE)-based controller is simulated. According to the results, not only is the performance of the proposed controller better under non-Gaussian situations but also effectively reaches the set point in the case of disturbance and uncertainty and provides higher control accuracy and robustness compared with the MSE-based MPC.