Comparison of control algorithms for the blood glucose concentration in a virtual patient with an artificial pancreas

Abstract

To obtain the most suitable control algorithm for a wearable artificial pancreas, different control algorithms were compared and tested using a Hovorka model. Model predictive control (MPC), linear and nonlinear model forms, proportional integral derivative control (PID), neural-network-based model predictive control (NN-MPC), nonlinear autoregressive moving average (NARMA-L2) and sequential quadratic programming (SQP) were evaluated using the Hovorka model. Due to the fact that modeling of biomedical processes are very complex, to present the most effective control algorithm, various control strategies were needed to application. In the control algorithms, set point tracking and disturbance rejection were performed. With respect to the rise times of the control algorithms, SQP with optimal control had the shortest time, and NARMA-L2 had the longest time. Because the control algorithm connects the glucose meter and the insulin pump in an artificial pancreas, the rise time is the most important parameter. We propose that optimal control with SQP is the most suitable control algorithm to connect the glucose meter and the insulin pump.