Design of Switched Model Predictive Control Algorithms for a Dual-Hormone Artificial Pancreas

In this paper, we evaluate the closed-loop performance of two switching strategies for a dual-hormone artificial pancreas (AP). The dual-hormone AP administers insulin and glucagon subcutaneously. Since insulin and glucagon have opposite effects, we want to avoid simultaneous injections of these two hormones. To handle non-simultaneous injections of insulin and glucagon, we compare model predictive control (MPC) algorithms using a hysteresis switch between insulin and glucagon controllers with a multiple-input single-output (MISO) formulation. Although the closed-loop performance of these two control strategies is similar, the hysteresis switch is preferable due to (i) its greater flexibility in control design and tuning and (ii) a more straightforward way to avoid simultaneous injections of insulin and glucagon.