Adaptive Nonlinear Model Predictive Control algorithm for blood glucose regulation in type 1 diabetic patients

Abstract

The pancreas of patients with Type 1 Diabetes Mellitus (T1DM) is unable to produce insulin. Thus, insulin therapy is required for T1DM to maintain Blood Glucose (BG) levels within the normal range. The Artificial Pancreas (AP) is a closed-loop control system that is used by T1D patients to maintain their BG levels at the normal range during daily life. In this work, an Adaptive Nonlinear Model Predictive Control (AMPC) algorithm for BG regulation in T1D patients is developed. The proposed technique uses the Feed Forward Neural Network (FFNN) as a nonlinear blood glucose prediction model to handle the delay between the moment of insulin injection and the moment of insulin interaction with the blood glucose. Also, it uses the Fuzzy Logic Controller (FLC) as a control algorithm to determine the amount of insulin required for regulating the BG level. An adaptation method is also included to adjust the proposed system to compensate for physiological differences among patients. The limits of the output membership functions for the FLC are optimized using the Genetic algorithm (GA). Simulation results for a 36h scenario are demonstrated in nine virtual adult patients. The master findings are the average percentages of these patients for the time spent in the normal range, hypo-, and hyperglycemia. Our results indicate that the proposed closed-loop control system increases the time that BG is in the normal range and causes less hyperglycemia as compared to a published technique studied in a similar scenario and population.